From 5274ec4dec498bd88ccbcd28862a0f78a3b95eff Mon Sep 17 00:00:00 2001 From: MerryMage Date: Thu, 28 Apr 2016 13:21:51 +0100 Subject: [PATCH] Stripped down version of SoundTouch 1.9.2 --- CMakeLists.txt | 14 + include/FIFOSampleBuffer.h | 184 ++++++ include/FIFOSamplePipe.h | 234 ++++++++ include/STTypes.h | 177 ++++++ include/SoundTouch.h | 301 ++++++++++ src/AAFilter.cpp | 236 ++++++++ src/AAFilter.h | 100 ++++ src/FIFOSampleBuffer.cpp | 274 +++++++++ src/FIRFilter.cpp | 328 +++++++++++ src/FIRFilter.h | 146 +++++ src/InterpolateLinear.cpp | 300 ++++++++++ src/InterpolateLinear.h | 92 +++ src/RateTransposer.cpp | 300 ++++++++++ src/RateTransposer.h | 179 ++++++ src/SoundTouch.cpp | 526 ++++++++++++++++++ src/TDStretch.cpp | 1078 ++++++++++++++++++++++++++++++++++++ src/TDStretch.h | 281 ++++++++++ src/cpu_detect.h | 62 +++ src/cpu_detect_x86.cpp | 138 +++++ 19 files changed, 4950 insertions(+) create mode 100644 CMakeLists.txt create mode 100644 include/FIFOSampleBuffer.h create mode 100644 include/FIFOSamplePipe.h create mode 100644 include/STTypes.h create mode 100644 include/SoundTouch.h create mode 100644 src/AAFilter.cpp create mode 100644 src/AAFilter.h create mode 100644 src/FIFOSampleBuffer.cpp create mode 100644 src/FIRFilter.cpp create mode 100644 src/FIRFilter.h create mode 100644 src/InterpolateLinear.cpp create mode 100644 src/InterpolateLinear.h create mode 100644 src/RateTransposer.cpp create mode 100644 src/RateTransposer.h create mode 100644 src/SoundTouch.cpp create mode 100644 src/TDStretch.cpp create mode 100644 src/TDStretch.h create mode 100644 src/cpu_detect.h create mode 100644 src/cpu_detect_x86.cpp diff --git a/CMakeLists.txt b/CMakeLists.txt new file mode 100644 index 0000000..30cc4e9 --- /dev/null +++ b/CMakeLists.txt @@ -0,0 +1,14 @@ +set(SRCS + src/AAFilter.cpp + src/cpu_detect_x86.cpp + src/FIFOSampleBuffer.cpp + src/FIRFilter.cpp + src/InterpolateLinear.cpp + src/RateTransposer.cpp + src/SoundTouch.cpp + src/TDStretch.cpp + ) + +include_directories(src include) + +add_library(SoundTouch STATIC ${SRCS}) diff --git a/include/FIFOSampleBuffer.h b/include/FIFOSampleBuffer.h new file mode 100644 index 0000000..4b22a51 --- /dev/null +++ b/include/FIFOSampleBuffer.h @@ -0,0 +1,184 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// A buffer class for temporarily storaging sound samples, operates as a +/// first-in-first-out pipe. +/// +/// Samples are added to the end of the sample buffer with the 'putSamples' +/// function, and are received from the beginning of the buffer by calling +/// the 'receiveSamples' function. The class automatically removes the +/// output samples from the buffer as well as grows the storage size +/// whenever necessary. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2014-01-05 23:40:22 +0200 (Sun, 05 Jan 2014) $ +// File revision : $Revision: 4 $ +// +// $Id: FIFOSampleBuffer.h 177 2014-01-05 21:40:22Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#ifndef FIFOSampleBuffer_H +#define FIFOSampleBuffer_H + +#include "FIFOSamplePipe.h" + +namespace soundtouch +{ + +/// Sample buffer working in FIFO (first-in-first-out) principle. The class takes +/// care of storage size adjustment and data moving during input/output operations. +/// +/// Notice that in case of stereo audio, one sample is considered to consist of +/// both channel data. +class FIFOSampleBuffer : public FIFOSamplePipe +{ +private: + /// Sample buffer. + SAMPLETYPE *buffer; + + // Raw unaligned buffer memory. 'buffer' is made aligned by pointing it to first + // 16-byte aligned location of this buffer + SAMPLETYPE *bufferUnaligned; + + /// Sample buffer size in bytes + uint sizeInBytes; + + /// How many samples are currently in buffer. + uint samplesInBuffer; + + /// Channels, 1=mono, 2=stereo. + uint channels; + + /// Current position pointer to the buffer. This pointer is increased when samples are + /// removed from the pipe so that it's necessary to actually rewind buffer (move data) + /// only new data when is put to the pipe. + uint bufferPos; + + /// Rewind the buffer by moving data from position pointed by 'bufferPos' to real + /// beginning of the buffer. + void rewind(); + + /// Ensures that the buffer has capacity for at least this many samples. + void ensureCapacity(uint capacityRequirement); + + /// Returns current capacity. + uint getCapacity() const; + +public: + + /// Constructor + FIFOSampleBuffer(int numChannels = 2 ///< Number of channels, 1=mono, 2=stereo. + ///< Default is stereo. + ); + + /// destructor + ~FIFOSampleBuffer(); + + /// Returns a pointer to the beginning of the output samples. + /// This function is provided for accessing the output samples directly. + /// Please be careful for not to corrupt the book-keeping! + /// + /// When using this function to output samples, also remember to 'remove' the + /// output samples from the buffer by calling the + /// 'receiveSamples(numSamples)' function + virtual SAMPLETYPE *ptrBegin(); + + /// Returns a pointer to the end of the used part of the sample buffer (i.e. + /// where the new samples are to be inserted). This function may be used for + /// inserting new samples into the sample buffer directly. Please be careful + /// not corrupt the book-keeping! + /// + /// When using this function as means for inserting new samples, also remember + /// to increase the sample count afterwards, by calling the + /// 'putSamples(numSamples)' function. + SAMPLETYPE *ptrEnd( + uint slackCapacity ///< How much free capacity (in samples) there _at least_ + ///< should be so that the caller can succesfully insert the + ///< desired samples to the buffer. If necessary, the function + ///< grows the buffer size to comply with this requirement. + ); + + /// Adds 'numSamples' pcs of samples from the 'samples' memory position to + /// the sample buffer. + virtual void putSamples(const SAMPLETYPE *samples, ///< Pointer to samples. + uint numSamples ///< Number of samples to insert. + ); + + /// Adjusts the book-keeping to increase number of samples in the buffer without + /// copying any actual samples. + /// + /// This function is used to update the number of samples in the sample buffer + /// when accessing the buffer directly with 'ptrEnd' function. Please be + /// careful though! + virtual void putSamples(uint numSamples ///< Number of samples been inserted. + ); + + /// Output samples from beginning of the sample buffer. Copies requested samples to + /// output buffer and removes them from the sample buffer. If there are less than + /// 'numsample' samples in the buffer, returns all that available. + /// + /// \return Number of samples returned. + virtual uint receiveSamples(SAMPLETYPE *output, ///< Buffer where to copy output samples. + uint maxSamples ///< How many samples to receive at max. + ); + + /// Adjusts book-keeping so that given number of samples are removed from beginning of the + /// sample buffer without copying them anywhere. + /// + /// Used to reduce the number of samples in the buffer when accessing the sample buffer directly + /// with 'ptrBegin' function. + virtual uint receiveSamples(uint maxSamples ///< Remove this many samples from the beginning of pipe. + ); + + /// Returns number of samples currently available. + virtual uint numSamples() const; + + /// Sets number of channels, 1 = mono, 2 = stereo. + void setChannels(int numChannels); + + /// Get number of channels + int getChannels() + { + return channels; + } + + /// Returns nonzero if there aren't any samples available for outputting. + virtual int isEmpty() const; + + /// Clears all the samples. + virtual void clear(); + + /// allow trimming (downwards) amount of samples in pipeline. + /// Returns adjusted amount of samples + uint adjustAmountOfSamples(uint numSamples); +}; + +} + +#endif diff --git a/include/FIFOSamplePipe.h b/include/FIFOSamplePipe.h new file mode 100644 index 0000000..f26c57b --- /dev/null +++ b/include/FIFOSamplePipe.h @@ -0,0 +1,234 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// 'FIFOSamplePipe' : An abstract base class for classes that manipulate sound +/// samples by operating like a first-in-first-out pipe: New samples are fed +/// into one end of the pipe with the 'putSamples' function, and the processed +/// samples are received from the other end with the 'receiveSamples' function. +/// +/// 'FIFOProcessor' : A base class for classes the do signal processing with +/// the samples while operating like a first-in-first-out pipe. When samples +/// are input with the 'putSamples' function, the class processes them +/// and moves the processed samples to the given 'output' pipe object, which +/// may be either another processing stage, or a fifo sample buffer object. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2012-06-13 22:29:53 +0300 (Wed, 13 Jun 2012) $ +// File revision : $Revision: 4 $ +// +// $Id: FIFOSamplePipe.h 143 2012-06-13 19:29:53Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#ifndef FIFOSamplePipe_H +#define FIFOSamplePipe_H + +#include +#include +#include "STTypes.h" + +namespace soundtouch +{ + +/// Abstract base class for FIFO (first-in-first-out) sample processing classes. +class FIFOSamplePipe +{ +public: + // virtual default destructor + virtual ~FIFOSamplePipe() {} + + + /// Returns a pointer to the beginning of the output samples. + /// This function is provided for accessing the output samples directly. + /// Please be careful for not to corrupt the book-keeping! + /// + /// When using this function to output samples, also remember to 'remove' the + /// output samples from the buffer by calling the + /// 'receiveSamples(numSamples)' function + virtual SAMPLETYPE *ptrBegin() = 0; + + /// Adds 'numSamples' pcs of samples from the 'samples' memory position to + /// the sample buffer. + virtual void putSamples(const SAMPLETYPE *samples, ///< Pointer to samples. + uint numSamples ///< Number of samples to insert. + ) = 0; + + + // Moves samples from the 'other' pipe instance to this instance. + void moveSamples(FIFOSamplePipe &other ///< Other pipe instance where from the receive the data. + ) + { + int oNumSamples = other.numSamples(); + + putSamples(other.ptrBegin(), oNumSamples); + other.receiveSamples(oNumSamples); + }; + + /// Output samples from beginning of the sample buffer. Copies requested samples to + /// output buffer and removes them from the sample buffer. If there are less than + /// 'numsample' samples in the buffer, returns all that available. + /// + /// \return Number of samples returned. + virtual uint receiveSamples(SAMPLETYPE *output, ///< Buffer where to copy output samples. + uint maxSamples ///< How many samples to receive at max. + ) = 0; + + /// Adjusts book-keeping so that given number of samples are removed from beginning of the + /// sample buffer without copying them anywhere. + /// + /// Used to reduce the number of samples in the buffer when accessing the sample buffer directly + /// with 'ptrBegin' function. + virtual uint receiveSamples(uint maxSamples ///< Remove this many samples from the beginning of pipe. + ) = 0; + + /// Returns number of samples currently available. + virtual uint numSamples() const = 0; + + // Returns nonzero if there aren't any samples available for outputting. + virtual int isEmpty() const = 0; + + /// Clears all the samples. + virtual void clear() = 0; + + /// allow trimming (downwards) amount of samples in pipeline. + /// Returns adjusted amount of samples + virtual uint adjustAmountOfSamples(uint numSamples) = 0; + +}; + + + +/// Base-class for sound processing routines working in FIFO principle. With this base +/// class it's easy to implement sound processing stages that can be chained together, +/// so that samples that are fed into beginning of the pipe automatically go through +/// all the processing stages. +/// +/// When samples are input to this class, they're first processed and then put to +/// the FIFO pipe that's defined as output of this class. This output pipe can be +/// either other processing stage or a FIFO sample buffer. +class FIFOProcessor :public FIFOSamplePipe +{ +protected: + /// Internal pipe where processed samples are put. + FIFOSamplePipe *output; + + /// Sets output pipe. + void setOutPipe(FIFOSamplePipe *pOutput) + { + assert(output == NULL); + assert(pOutput != NULL); + output = pOutput; + } + + + /// Constructor. Doesn't define output pipe; it has to be set be + /// 'setOutPipe' function. + FIFOProcessor() + { + output = NULL; + } + + + /// Constructor. Configures output pipe. + FIFOProcessor(FIFOSamplePipe *pOutput ///< Output pipe. + ) + { + output = pOutput; + } + + + /// Destructor. + virtual ~FIFOProcessor() + { + } + + + /// Returns a pointer to the beginning of the output samples. + /// This function is provided for accessing the output samples directly. + /// Please be careful for not to corrupt the book-keeping! + /// + /// When using this function to output samples, also remember to 'remove' the + /// output samples from the buffer by calling the + /// 'receiveSamples(numSamples)' function + virtual SAMPLETYPE *ptrBegin() + { + return output->ptrBegin(); + } + +public: + + /// Output samples from beginning of the sample buffer. Copies requested samples to + /// output buffer and removes them from the sample buffer. If there are less than + /// 'numsample' samples in the buffer, returns all that available. + /// + /// \return Number of samples returned. + virtual uint receiveSamples(SAMPLETYPE *outBuffer, ///< Buffer where to copy output samples. + uint maxSamples ///< How many samples to receive at max. + ) + { + return output->receiveSamples(outBuffer, maxSamples); + } + + + /// Adjusts book-keeping so that given number of samples are removed from beginning of the + /// sample buffer without copying them anywhere. + /// + /// Used to reduce the number of samples in the buffer when accessing the sample buffer directly + /// with 'ptrBegin' function. + virtual uint receiveSamples(uint maxSamples ///< Remove this many samples from the beginning of pipe. + ) + { + return output->receiveSamples(maxSamples); + } + + + /// Returns number of samples currently available. + virtual uint numSamples() const + { + return output->numSamples(); + } + + + /// Returns nonzero if there aren't any samples available for outputting. + virtual int isEmpty() const + { + return output->isEmpty(); + } + + /// allow trimming (downwards) amount of samples in pipeline. + /// Returns adjusted amount of samples + virtual uint adjustAmountOfSamples(uint numSamples) + { + return output->adjustAmountOfSamples(numSamples); + } + +}; + +} + +#endif diff --git a/include/STTypes.h b/include/STTypes.h new file mode 100644 index 0000000..b257974 --- /dev/null +++ b/include/STTypes.h @@ -0,0 +1,177 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// Common type definitions for SoundTouch audio processing library. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2015-05-18 18:25:07 +0300 (Mon, 18 May 2015) $ +// File revision : $Revision: 3 $ +// +// $Id: STTypes.h 215 2015-05-18 15:25:07Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#ifndef STTypes_H +#define STTypes_H + +typedef unsigned int uint; +typedef unsigned long ulong; + +// Patch for MinGW: on Win64 long is 32-bit +#ifdef _WIN64 + typedef unsigned long long ulongptr; +#else + typedef ulong ulongptr; +#endif + + +// Helper macro for aligning pointer up to next 16-byte boundary +#define SOUNDTOUCH_ALIGN_POINTER_16(x) ( ( (ulongptr)(x) + 15 ) & ~(ulongptr)15 ) + +namespace soundtouch +{ + /// Activate these undef's to overrule the possible sampletype + /// setting inherited from some other header file: + //#undef SOUNDTOUCH_INTEGER_SAMPLES + //#undef SOUNDTOUCH_FLOAT_SAMPLES + + /// If following flag is defined, always uses multichannel processing + /// routines also for mono and stero sound. This is for routine testing + /// purposes; output should be same with either routines, yet disabling + /// the dedicated mono/stereo processing routines will result in slower + /// runtime performance so recommendation is to keep this off. + // #define USE_MULTICH_ALWAYS + + #if (defined(__SOFTFP__) && defined(ANDROID)) + // For Android compilation: Force use of Integer samples in case that + // compilation uses soft-floating point emulation - soft-fp is way too slow + #undef SOUNDTOUCH_FLOAT_SAMPLES + #define SOUNDTOUCH_INTEGER_SAMPLES 1 + #endif + + #if !(SOUNDTOUCH_INTEGER_SAMPLES || SOUNDTOUCH_FLOAT_SAMPLES) + + /// Choose either 32bit floating point or 16bit integer sampletype + /// by choosing one of the following defines, unless this selection + /// has already been done in some other file. + //// + /// Notes: + /// - In Windows environment, choose the sample format with the + /// following defines. + /// - In GNU environment, the floating point samples are used by + /// default, but integer samples can be chosen by giving the + /// following switch to the configure script: + /// ./configure --enable-integer-samples + /// However, if you still prefer to select the sample format here + /// also in GNU environment, then please #undef the INTEGER_SAMPLE + /// and FLOAT_SAMPLE defines first as in comments above. + #define SOUNDTOUCH_INTEGER_SAMPLES 1 //< 16bit integer samples + //#define SOUNDTOUCH_FLOAT_SAMPLES 1 //< 32bit float samples + + #endif + + #if (_M_IX86 || __i386__ || __x86_64__ || _M_X64) + /// Define this to allow X86-specific assembler/intrinsic optimizations. + /// Notice that library contains also usual C++ versions of each of these + /// these routines, so if you're having difficulties getting the optimized + /// routines compiled for whatever reason, you may disable these optimizations + /// to make the library compile. + + //#define SOUNDTOUCH_ALLOW_X86_OPTIMIZATIONS 1 + + /// In GNU environment, allow the user to override this setting by + /// giving the following switch to the configure script: + /// ./configure --disable-x86-optimizations + /// ./configure --enable-x86-optimizations=no + #ifdef SOUNDTOUCH_DISABLE_X86_OPTIMIZATIONS + #undef SOUNDTOUCH_ALLOW_X86_OPTIMIZATIONS + #endif + #else + /// Always disable optimizations when not using a x86 systems. + #undef SOUNDTOUCH_ALLOW_X86_OPTIMIZATIONS + + #endif + + // If defined, allows the SIMD-optimized routines to take minor shortcuts + // for improved performance. Undefine to require faithfully similar SIMD + // calculations as in normal C implementation. + #define SOUNDTOUCH_ALLOW_NONEXACT_SIMD_OPTIMIZATION 1 + + + #ifdef SOUNDTOUCH_INTEGER_SAMPLES + // 16bit integer sample type + typedef short SAMPLETYPE; + // data type for sample accumulation: Use 32bit integer to prevent overflows + typedef long LONG_SAMPLETYPE; + + #ifdef SOUNDTOUCH_FLOAT_SAMPLES + // check that only one sample type is defined + #error "conflicting sample types defined" + #endif // SOUNDTOUCH_FLOAT_SAMPLES + + #ifdef SOUNDTOUCH_ALLOW_X86_OPTIMIZATIONS + // Allow MMX optimizations + #define SOUNDTOUCH_ALLOW_MMX 1 + #endif + + #else + + // floating point samples + typedef float SAMPLETYPE; + // data type for sample accumulation: Use double to utilize full precision. + typedef double LONG_SAMPLETYPE; + + #ifdef SOUNDTOUCH_ALLOW_X86_OPTIMIZATIONS + // Allow SSE optimizations + #define SOUNDTOUCH_ALLOW_SSE 1 + #endif + + #endif // SOUNDTOUCH_INTEGER_SAMPLES + +}; + +// define ST_NO_EXCEPTION_HANDLING switch to disable throwing std exceptions: +// #define ST_NO_EXCEPTION_HANDLING 1 +#ifdef ST_NO_EXCEPTION_HANDLING + // Exceptions disabled. Throw asserts instead if enabled. + #include + #define ST_THROW_RT_ERROR(x) {assert((const char *)x);} +#else + // use c++ standard exceptions + #include + #include + #define ST_THROW_RT_ERROR(x) {throw std::runtime_error(x);} +#endif + +// When this #define is active, eliminates a clicking sound when the "rate" or "pitch" +// parameter setting crosses from value <1 to >=1 or vice versa during processing. +// Default is off as such crossover is untypical case and involves a slight sound +// quality compromise. +//#define SOUNDTOUCH_PREVENT_CLICK_AT_RATE_CROSSOVER 1 + +#endif diff --git a/include/SoundTouch.h b/include/SoundTouch.h new file mode 100644 index 0000000..b03855e --- /dev/null +++ b/include/SoundTouch.h @@ -0,0 +1,301 @@ +////////////////////////////////////////////////////////////////////////////// +/// +/// SoundTouch - main class for tempo/pitch/rate adjusting routines. +/// +/// Notes: +/// - Initialize the SoundTouch object instance by setting up the sound stream +/// parameters with functions 'setSampleRate' and 'setChannels', then set +/// desired tempo/pitch/rate settings with the corresponding functions. +/// +/// - The SoundTouch class behaves like a first-in-first-out pipeline: The +/// samples that are to be processed are fed into one of the pipe by calling +/// function 'putSamples', while the ready processed samples can be read +/// from the other end of the pipeline with function 'receiveSamples'. +/// +/// - The SoundTouch processing classes require certain sized 'batches' of +/// samples in order to process the sound. For this reason the classes buffer +/// incoming samples until there are enough of samples available for +/// processing, then they carry out the processing step and consequently +/// make the processed samples available for outputting. +/// +/// - For the above reason, the processing routines introduce a certain +/// 'latency' between the input and output, so that the samples input to +/// SoundTouch may not be immediately available in the output, and neither +/// the amount of outputtable samples may not immediately be in direct +/// relationship with the amount of previously input samples. +/// +/// - The tempo/pitch/rate control parameters can be altered during processing. +/// Please notice though that they aren't currently protected by semaphores, +/// so in multi-thread application external semaphore protection may be +/// required. +/// +/// - This class utilizes classes 'TDStretch' for tempo change (without modifying +/// pitch) and 'RateTransposer' for changing the playback rate (that is, both +/// tempo and pitch in the same ratio) of the sound. The third available control +/// 'pitch' (change pitch but maintain tempo) is produced by a combination of +/// combining the two other controls. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2015-09-20 10:38:32 +0300 (Sun, 20 Sep 2015) $ +// File revision : $Revision: 4 $ +// +// $Id: SoundTouch.h 230 2015-09-20 07:38:32Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#ifndef SoundTouch_H +#define SoundTouch_H + +#include "FIFOSamplePipe.h" +#include "STTypes.h" + +namespace soundtouch +{ + +/// Soundtouch library version string +#define SOUNDTOUCH_VERSION "1.9.2" + +/// SoundTouch library version id +#define SOUNDTOUCH_VERSION_ID (10902) + +// +// Available setting IDs for the 'setSetting' & 'get_setting' functions: + +/// Enable/disable anti-alias filter in pitch transposer (0 = disable) +#define SETTING_USE_AA_FILTER 0 + +/// Pitch transposer anti-alias filter length (8 .. 128 taps, default = 32) +#define SETTING_AA_FILTER_LENGTH 1 + +/// Enable/disable quick seeking algorithm in tempo changer routine +/// (enabling quick seeking lowers CPU utilization but causes a minor sound +/// quality compromising) +#define SETTING_USE_QUICKSEEK 2 + +/// Time-stretch algorithm single processing sequence length in milliseconds. This determines +/// to how long sequences the original sound is chopped in the time-stretch algorithm. +/// See "STTypes.h" or README for more information. +#define SETTING_SEQUENCE_MS 3 + +/// Time-stretch algorithm seeking window length in milliseconds for algorithm that finds the +/// best possible overlapping location. This determines from how wide window the algorithm +/// may look for an optimal joining location when mixing the sound sequences back together. +/// See "STTypes.h" or README for more information. +#define SETTING_SEEKWINDOW_MS 4 + +/// Time-stretch algorithm overlap length in milliseconds. When the chopped sound sequences +/// are mixed back together, to form a continuous sound stream, this parameter defines over +/// how long period the two consecutive sequences are let to overlap each other. +/// See "STTypes.h" or README for more information. +#define SETTING_OVERLAP_MS 5 + + +/// Call "getSetting" with this ID to query nominal average processing sequence +/// size in samples. This value tells approcimate value how many input samples +/// SoundTouch needs to gather before it does DSP processing run for the sample batch. +/// +/// Notices: +/// - This is read-only parameter, i.e. setSetting ignores this parameter +/// - Returned value is approximate average value, exact processing batch +/// size may wary from time to time +/// - This parameter value is not constant but may change depending on +/// tempo/pitch/rate/samplerate settings. +#define SETTING_NOMINAL_INPUT_SEQUENCE 6 + + +/// Call "getSetting" with this ID to query nominal average processing output +/// size in samples. This value tells approcimate value how many output samples +/// SoundTouch outputs once it does DSP processing run for a batch of input samples. +/// +/// Notices: +/// - This is read-only parameter, i.e. setSetting ignores this parameter +/// - Returned value is approximate average value, exact processing batch +/// size may wary from time to time +/// - This parameter value is not constant but may change depending on +/// tempo/pitch/rate/samplerate settings. +#define SETTING_NOMINAL_OUTPUT_SEQUENCE 7 + +class SoundTouch : public FIFOProcessor +{ +private: + /// Rate transposer class instance + class RateTransposer *pRateTransposer; + + /// Time-stretch class instance + class TDStretch *pTDStretch; + + /// Virtual pitch parameter. Effective rate & tempo are calculated from these parameters. + double virtualRate; + + /// Virtual pitch parameter. Effective rate & tempo are calculated from these parameters. + double virtualTempo; + + /// Virtual pitch parameter. Effective rate & tempo are calculated from these parameters. + double virtualPitch; + + /// Flag: Has sample rate been set? + bool bSrateSet; + + /// Accumulator for how many samples in total will be expected as output vs. samples put in, + /// considering current processing settings. + double samplesExpectedOut; + + /// Accumulator for how many samples in total have been read out from the processing so far + long samplesOutput; + + /// Calculates effective rate & tempo valuescfrom 'virtualRate', 'virtualTempo' and + /// 'virtualPitch' parameters. + void calcEffectiveRateAndTempo(); + +protected : + /// Number of channels + uint channels; + + /// Effective 'rate' value calculated from 'virtualRate', 'virtualTempo' and 'virtualPitch' + double rate; + + /// Effective 'tempo' value calculated from 'virtualRate', 'virtualTempo' and 'virtualPitch' + double tempo; + +public: + SoundTouch(); + virtual ~SoundTouch(); + + /// Get SoundTouch library version string + static const char *getVersionString(); + + /// Get SoundTouch library version Id + static uint getVersionId(); + + /// Sets new rate control value. Normal rate = 1.0, smaller values + /// represent slower rate, larger faster rates. + void setRate(double newRate); + + /// Sets new tempo control value. Normal tempo = 1.0, smaller values + /// represent slower tempo, larger faster tempo. + void setTempo(double newTempo); + + /// Sets new rate control value as a difference in percents compared + /// to the original rate (-50 .. +100 %) + void setRateChange(double newRate); + + /// Sets new tempo control value as a difference in percents compared + /// to the original tempo (-50 .. +100 %) + void setTempoChange(double newTempo); + + /// Sets new pitch control value. Original pitch = 1.0, smaller values + /// represent lower pitches, larger values higher pitch. + void setPitch(double newPitch); + + /// Sets pitch change in octaves compared to the original pitch + /// (-1.00 .. +1.00) + void setPitchOctaves(double newPitch); + + /// Sets pitch change in semi-tones compared to the original pitch + /// (-12 .. +12) + void setPitchSemiTones(int newPitch); + void setPitchSemiTones(double newPitch); + + /// Sets the number of channels, 1 = mono, 2 = stereo + void setChannels(uint numChannels); + + /// Sets sample rate. + void setSampleRate(uint srate); + + /// Flushes the last samples from the processing pipeline to the output. + /// Clears also the internal processing buffers. + // + /// Note: This function is meant for extracting the last samples of a sound + /// stream. This function may introduce additional blank samples in the end + /// of the sound stream, and thus it's not recommended to call this function + /// in the middle of a sound stream. + void flush(); + + /// Adds 'numSamples' pcs of samples from the 'samples' memory position into + /// the input of the object. Notice that sample rate _has_to_ be set before + /// calling this function, otherwise throws a runtime_error exception. + virtual void putSamples( + const SAMPLETYPE *samples, ///< Pointer to sample buffer. + uint numSamples ///< Number of samples in buffer. Notice + ///< that in case of stereo-sound a single sample + ///< contains data for both channels. + ); + + /// Output samples from beginning of the sample buffer. Copies requested samples to + /// output buffer and removes them from the sample buffer. If there are less than + /// 'numsample' samples in the buffer, returns all that available. + /// + /// \return Number of samples returned. + virtual uint receiveSamples(SAMPLETYPE *output, ///< Buffer where to copy output samples. + uint maxSamples ///< How many samples to receive at max. + ); + + /// Adjusts book-keeping so that given number of samples are removed from beginning of the + /// sample buffer without copying them anywhere. + /// + /// Used to reduce the number of samples in the buffer when accessing the sample buffer directly + /// with 'ptrBegin' function. + virtual uint receiveSamples(uint maxSamples ///< Remove this many samples from the beginning of pipe. + ); + + /// Clears all the samples in the object's output and internal processing + /// buffers. + virtual void clear(); + + /// Changes a setting controlling the processing system behaviour. See the + /// 'SETTING_...' defines for available setting ID's. + /// + /// \return 'true' if the setting was succesfully changed + bool setSetting(int settingId, ///< Setting ID number. see SETTING_... defines. + int value ///< New setting value. + ); + + /// Reads a setting controlling the processing system behaviour. See the + /// 'SETTING_...' defines for available setting ID's. + /// + /// \return the setting value. + int getSetting(int settingId ///< Setting ID number, see SETTING_... defines. + ) const; + + /// Returns number of samples currently unprocessed. + virtual uint numUnprocessedSamples() const; + + + /// Other handy functions that are implemented in the ancestor classes (see + /// classes 'FIFOProcessor' and 'FIFOSamplePipe') + /// + /// - receiveSamples() : Use this function to receive 'ready' processed samples from SoundTouch. + /// - numSamples() : Get number of 'ready' samples that can be received with + /// function 'receiveSamples()' + /// - isEmpty() : Returns nonzero if there aren't any 'ready' samples. + /// - clear() : Clears all samples from ready/processing buffers. +}; + +} +#endif diff --git a/src/AAFilter.cpp b/src/AAFilter.cpp new file mode 100644 index 0000000..a254300 --- /dev/null +++ b/src/AAFilter.cpp @@ -0,0 +1,236 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// FIR low-pass (anti-alias) filter with filter coefficient design routine and +/// MMX optimization. +/// +/// Anti-alias filter is used to prevent folding of high frequencies when +/// transposing the sample rate with interpolation. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2014-01-05 23:40:22 +0200 (Sun, 05 Jan 2014) $ +// File revision : $Revision: 4 $ +// +// $Id: AAFilter.cpp 177 2014-01-05 21:40:22Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#include +#include +#include +#include +#include "AAFilter.h" +#include "FIRFilter.h" + +using namespace soundtouch; + +#define PI 3.141592655357989 +#define TWOPI (2 * PI) + +// define this to save AA filter coefficients to a file +// #define _DEBUG_SAVE_AAFILTER_COEFFICIENTS 1 + +#ifdef _DEBUG_SAVE_AAFILTER_COEFFICIENTS + #include + + static void _DEBUG_SAVE_AAFIR_COEFFS(SAMPLETYPE *coeffs, int len) + { + FILE *fptr = fopen("aa_filter_coeffs.txt", "wt"); + if (fptr == NULL) return; + + for (int i = 0; i < len; i ++) + { + double temp = coeffs[i]; + fprintf(fptr, "%lf\n", temp); + } + fclose(fptr); + } + +#else + #define _DEBUG_SAVE_AAFIR_COEFFS(x, y) +#endif + + +/***************************************************************************** + * + * Implementation of the class 'AAFilter' + * + *****************************************************************************/ + +AAFilter::AAFilter(uint len) +{ + pFIR = FIRFilter::newInstance(); + cutoffFreq = 0.5; + setLength(len); +} + + + +AAFilter::~AAFilter() +{ + delete pFIR; +} + + + +// Sets new anti-alias filter cut-off edge frequency, scaled to +// sampling frequency (nyquist frequency = 0.5). +// The filter will cut frequencies higher than the given frequency. +void AAFilter::setCutoffFreq(double newCutoffFreq) +{ + cutoffFreq = newCutoffFreq; + calculateCoeffs(); +} + + + +// Sets number of FIR filter taps +void AAFilter::setLength(uint newLength) +{ + length = newLength; + calculateCoeffs(); +} + + + +// Calculates coefficients for a low-pass FIR filter using Hamming window +void AAFilter::calculateCoeffs() +{ + uint i; + double cntTemp, temp, tempCoeff,h, w; + double wc; + double scaleCoeff, sum; + double *work; + SAMPLETYPE *coeffs; + + assert(length >= 2); + assert(length % 4 == 0); + assert(cutoffFreq >= 0); + assert(cutoffFreq <= 0.5); + + work = new double[length]; + coeffs = new SAMPLETYPE[length]; + + wc = 2.0 * PI * cutoffFreq; + tempCoeff = TWOPI / (double)length; + + sum = 0; + for (i = 0; i < length; i ++) + { + cntTemp = (double)i - (double)(length / 2); + + temp = cntTemp * wc; + if (temp != 0) + { + h = sin(temp) / temp; // sinc function + } + else + { + h = 1.0; + } + w = 0.54 + 0.46 * cos(tempCoeff * cntTemp); // hamming window + + temp = w * h; + work[i] = temp; + + // calc net sum of coefficients + sum += temp; + } + + // ensure the sum of coefficients is larger than zero + assert(sum > 0); + + // ensure we've really designed a lowpass filter... + assert(work[length/2] > 0); + assert(work[length/2 + 1] > -1e-6); + assert(work[length/2 - 1] > -1e-6); + + // Calculate a scaling coefficient in such a way that the result can be + // divided by 16384 + scaleCoeff = 16384.0f / sum; + + for (i = 0; i < length; i ++) + { + temp = work[i] * scaleCoeff; +//#if SOUNDTOUCH_INTEGER_SAMPLES + // scale & round to nearest integer + temp += (temp >= 0) ? 0.5 : -0.5; + // ensure no overfloods + assert(temp >= -32768 && temp <= 32767); +//#endif + coeffs[i] = (SAMPLETYPE)temp; + } + + // Set coefficients. Use divide factor 14 => divide result by 2^14 = 16384 + pFIR->setCoefficients(coeffs, length, 14); + + _DEBUG_SAVE_AAFIR_COEFFS(coeffs, length); + + delete[] work; + delete[] coeffs; +} + + +// Applies the filter to the given sequence of samples. +// Note : The amount of outputted samples is by value of 'filter length' +// smaller than the amount of input samples. +uint AAFilter::evaluate(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples, uint numChannels) const +{ + return pFIR->evaluate(dest, src, numSamples, numChannels); +} + + +/// Applies the filter to the given src & dest pipes, so that processed amount of +/// samples get removed from src, and produced amount added to dest +/// Note : The amount of outputted samples is by value of 'filter length' +/// smaller than the amount of input samples. +uint AAFilter::evaluate(FIFOSampleBuffer &dest, FIFOSampleBuffer &src) const +{ + SAMPLETYPE *pdest; + const SAMPLETYPE *psrc; + uint numSrcSamples; + uint result; + int numChannels = src.getChannels(); + + assert(numChannels == dest.getChannels()); + + numSrcSamples = src.numSamples(); + psrc = src.ptrBegin(); + pdest = dest.ptrEnd(numSrcSamples); + result = pFIR->evaluate(pdest, psrc, numSrcSamples, numChannels); + src.receiveSamples(result); + dest.putSamples(result); + + return result; +} + + +uint AAFilter::getLength() const +{ + return pFIR->getLength(); +} diff --git a/src/AAFilter.h b/src/AAFilter.h new file mode 100644 index 0000000..8c34b6b --- /dev/null +++ b/src/AAFilter.h @@ -0,0 +1,100 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// Sampled sound tempo changer/time stretch algorithm. Changes the sound tempo +/// while maintaining the original pitch by using a time domain WSOLA-like method +/// with several performance-increasing tweaks. +/// +/// Anti-alias filter is used to prevent folding of high frequencies when +/// transposing the sample rate with interpolation. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2014-01-07 21:41:23 +0200 (Tue, 07 Jan 2014) $ +// File revision : $Revision: 4 $ +// +// $Id: AAFilter.h 187 2014-01-07 19:41:23Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#ifndef AAFilter_H +#define AAFilter_H + +#include "STTypes.h" +#include "FIFOSampleBuffer.h" + +namespace soundtouch +{ + +class AAFilter +{ +protected: + class FIRFilter *pFIR; + + /// Low-pass filter cut-off frequency, negative = invalid + double cutoffFreq; + + /// num of filter taps + uint length; + + /// Calculate the FIR coefficients realizing the given cutoff-frequency + void calculateCoeffs(); +public: + AAFilter(uint length); + + ~AAFilter(); + + /// Sets new anti-alias filter cut-off edge frequency, scaled to sampling + /// frequency (nyquist frequency = 0.5). The filter will cut off the + /// frequencies than that. + void setCutoffFreq(double newCutoffFreq); + + /// Sets number of FIR filter taps, i.e. ~filter complexity + void setLength(uint newLength); + + uint getLength() const; + + /// Applies the filter to the given sequence of samples. + /// Note : The amount of outputted samples is by value of 'filter length' + /// smaller than the amount of input samples. + uint evaluate(SAMPLETYPE *dest, + const SAMPLETYPE *src, + uint numSamples, + uint numChannels) const; + + /// Applies the filter to the given src & dest pipes, so that processed amount of + /// samples get removed from src, and produced amount added to dest + /// Note : The amount of outputted samples is by value of 'filter length' + /// smaller than the amount of input samples. + uint evaluate(FIFOSampleBuffer &dest, + FIFOSampleBuffer &src) const; + +}; + +} + +#endif diff --git a/src/FIFOSampleBuffer.cpp b/src/FIFOSampleBuffer.cpp new file mode 100644 index 0000000..4d9740a --- /dev/null +++ b/src/FIFOSampleBuffer.cpp @@ -0,0 +1,274 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// A buffer class for temporarily storaging sound samples, operates as a +/// first-in-first-out pipe. +/// +/// Samples are added to the end of the sample buffer with the 'putSamples' +/// function, and are received from the beginning of the buffer by calling +/// the 'receiveSamples' function. The class automatically removes the +/// outputted samples from the buffer, as well as grows the buffer size +/// whenever necessary. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2012-11-08 20:53:01 +0200 (Thu, 08 Nov 2012) $ +// File revision : $Revision: 4 $ +// +// $Id: FIFOSampleBuffer.cpp 160 2012-11-08 18:53:01Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#include +#include +#include +#include + +#include "FIFOSampleBuffer.h" + +using namespace soundtouch; + +// Constructor +FIFOSampleBuffer::FIFOSampleBuffer(int numChannels) +{ + assert(numChannels > 0); + sizeInBytes = 0; // reasonable initial value + buffer = NULL; + bufferUnaligned = NULL; + samplesInBuffer = 0; + bufferPos = 0; + channels = (uint)numChannels; + ensureCapacity(32); // allocate initial capacity +} + + +// destructor +FIFOSampleBuffer::~FIFOSampleBuffer() +{ + delete[] bufferUnaligned; + bufferUnaligned = NULL; + buffer = NULL; +} + + +// Sets number of channels, 1 = mono, 2 = stereo +void FIFOSampleBuffer::setChannels(int numChannels) +{ + uint usedBytes; + + assert(numChannels > 0); + usedBytes = channels * samplesInBuffer; + channels = (uint)numChannels; + samplesInBuffer = usedBytes / channels; +} + + +// if output location pointer 'bufferPos' isn't zero, 'rewinds' the buffer and +// zeroes this pointer by copying samples from the 'bufferPos' pointer +// location on to the beginning of the buffer. +void FIFOSampleBuffer::rewind() +{ + if (buffer && bufferPos) + { + memmove(buffer, ptrBegin(), sizeof(SAMPLETYPE) * channels * samplesInBuffer); + bufferPos = 0; + } +} + + +// Adds 'numSamples' pcs of samples from the 'samples' memory position to +// the sample buffer. +void FIFOSampleBuffer::putSamples(const SAMPLETYPE *samples, uint nSamples) +{ + memcpy(ptrEnd(nSamples), samples, sizeof(SAMPLETYPE) * nSamples * channels); + samplesInBuffer += nSamples; +} + + +// Increases the number of samples in the buffer without copying any actual +// samples. +// +// This function is used to update the number of samples in the sample buffer +// when accessing the buffer directly with 'ptrEnd' function. Please be +// careful though! +void FIFOSampleBuffer::putSamples(uint nSamples) +{ + uint req; + + req = samplesInBuffer + nSamples; + ensureCapacity(req); + samplesInBuffer += nSamples; +} + + +// Returns a pointer to the end of the used part of the sample buffer (i.e. +// where the new samples are to be inserted). This function may be used for +// inserting new samples into the sample buffer directly. Please be careful! +// +// Parameter 'slackCapacity' tells the function how much free capacity (in +// terms of samples) there _at least_ should be, in order to the caller to +// succesfully insert all the required samples to the buffer. When necessary, +// the function grows the buffer size to comply with this requirement. +// +// When using this function as means for inserting new samples, also remember +// to increase the sample count afterwards, by calling the +// 'putSamples(numSamples)' function. +SAMPLETYPE *FIFOSampleBuffer::ptrEnd(uint slackCapacity) +{ + ensureCapacity(samplesInBuffer + slackCapacity); + return buffer + samplesInBuffer * channels; +} + + +// Returns a pointer to the beginning of the currently non-outputted samples. +// This function is provided for accessing the output samples directly. +// Please be careful! +// +// When using this function to output samples, also remember to 'remove' the +// outputted samples from the buffer by calling the +// 'receiveSamples(numSamples)' function +SAMPLETYPE *FIFOSampleBuffer::ptrBegin() +{ + assert(buffer); + return buffer + bufferPos * channels; +} + + +// Ensures that the buffer has enought capacity, i.e. space for _at least_ +// 'capacityRequirement' number of samples. The buffer is grown in steps of +// 4 kilobytes to eliminate the need for frequently growing up the buffer, +// as well as to round the buffer size up to the virtual memory page size. +void FIFOSampleBuffer::ensureCapacity(uint capacityRequirement) +{ + SAMPLETYPE *tempUnaligned, *temp; + + if (capacityRequirement > getCapacity()) + { + // enlarge the buffer in 4kbyte steps (round up to next 4k boundary) + sizeInBytes = (capacityRequirement * channels * sizeof(SAMPLETYPE) + 4095) & (uint)-4096; + assert(sizeInBytes % 2 == 0); + tempUnaligned = new SAMPLETYPE[sizeInBytes / sizeof(SAMPLETYPE) + 16 / sizeof(SAMPLETYPE)]; + if (tempUnaligned == NULL) + { + ST_THROW_RT_ERROR("Couldn't allocate memory!\n"); + } + // Align the buffer to begin at 16byte cache line boundary for optimal performance + temp = (SAMPLETYPE *)SOUNDTOUCH_ALIGN_POINTER_16(tempUnaligned); + if (samplesInBuffer) + { + memcpy(temp, ptrBegin(), samplesInBuffer * channels * sizeof(SAMPLETYPE)); + } + delete[] bufferUnaligned; + buffer = temp; + bufferUnaligned = tempUnaligned; + bufferPos = 0; + } + else + { + // simply rewind the buffer (if necessary) + rewind(); + } +} + + +// Returns the current buffer capacity in terms of samples +uint FIFOSampleBuffer::getCapacity() const +{ + return sizeInBytes / (channels * sizeof(SAMPLETYPE)); +} + + +// Returns the number of samples currently in the buffer +uint FIFOSampleBuffer::numSamples() const +{ + return samplesInBuffer; +} + + +// Output samples from beginning of the sample buffer. Copies demanded number +// of samples to output and removes them from the sample buffer. If there +// are less than 'numsample' samples in the buffer, returns all available. +// +// Returns number of samples copied. +uint FIFOSampleBuffer::receiveSamples(SAMPLETYPE *output, uint maxSamples) +{ + uint num; + + num = (maxSamples > samplesInBuffer) ? samplesInBuffer : maxSamples; + + memcpy(output, ptrBegin(), channels * sizeof(SAMPLETYPE) * num); + return receiveSamples(num); +} + + +// Removes samples from the beginning of the sample buffer without copying them +// anywhere. Used to reduce the number of samples in the buffer, when accessing +// the sample buffer with the 'ptrBegin' function. +uint FIFOSampleBuffer::receiveSamples(uint maxSamples) +{ + if (maxSamples >= samplesInBuffer) + { + uint temp; + + temp = samplesInBuffer; + samplesInBuffer = 0; + return temp; + } + + samplesInBuffer -= maxSamples; + bufferPos += maxSamples; + + return maxSamples; +} + + +// Returns nonzero if the sample buffer is empty +int FIFOSampleBuffer::isEmpty() const +{ + return (samplesInBuffer == 0) ? 1 : 0; +} + + +// Clears the sample buffer +void FIFOSampleBuffer::clear() +{ + samplesInBuffer = 0; + bufferPos = 0; +} + + +/// allow trimming (downwards) amount of samples in pipeline. +/// Returns adjusted amount of samples +uint FIFOSampleBuffer::adjustAmountOfSamples(uint numSamples) +{ + if (numSamples < samplesInBuffer) + { + samplesInBuffer = numSamples; + } + return samplesInBuffer; +} + diff --git a/src/FIRFilter.cpp b/src/FIRFilter.cpp new file mode 100644 index 0000000..456418a --- /dev/null +++ b/src/FIRFilter.cpp @@ -0,0 +1,328 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// General FIR digital filter routines with MMX optimization. +/// +/// Note : MMX optimized functions reside in a separate, platform-specific file, +/// e.g. 'mmx_win.cpp' or 'mmx_gcc.cpp' +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2015-02-21 23:24:29 +0200 (Sat, 21 Feb 2015) $ +// File revision : $Revision: 4 $ +// +// $Id: FIRFilter.cpp 202 2015-02-21 21:24:29Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#include +#include +#include +#include +#include "FIRFilter.h" +#include "cpu_detect.h" + +using namespace soundtouch; + +/***************************************************************************** + * + * Implementation of the class 'FIRFilter' + * + *****************************************************************************/ + +FIRFilter::FIRFilter() +{ + resultDivFactor = 0; + resultDivider = 0; + length = 0; + lengthDiv8 = 0; + filterCoeffs = NULL; +} + + +FIRFilter::~FIRFilter() +{ + delete[] filterCoeffs; +} + +// Usual C-version of the filter routine for stereo sound +uint FIRFilter::evaluateFilterStereo(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples) const +{ + int j, end; +#ifdef SOUNDTOUCH_FLOAT_SAMPLES + // when using floating point samples, use a scaler instead of a divider + // because division is much slower operation than multiplying. + double dScaler = 1.0 / (double)resultDivider; +#endif + + assert(length != 0); + assert(src != NULL); + assert(dest != NULL); + assert(filterCoeffs != NULL); + + end = 2 * (numSamples - length); + + #pragma omp parallel for + for (j = 0; j < end; j += 2) + { + const SAMPLETYPE *ptr; + LONG_SAMPLETYPE suml, sumr; + uint i; + + suml = sumr = 0; + ptr = src + j; + + for (i = 0; i < length; i += 4) + { + // loop is unrolled by factor of 4 here for efficiency + suml += ptr[2 * i + 0] * filterCoeffs[i + 0] + + ptr[2 * i + 2] * filterCoeffs[i + 1] + + ptr[2 * i + 4] * filterCoeffs[i + 2] + + ptr[2 * i + 6] * filterCoeffs[i + 3]; + sumr += ptr[2 * i + 1] * filterCoeffs[i + 0] + + ptr[2 * i + 3] * filterCoeffs[i + 1] + + ptr[2 * i + 5] * filterCoeffs[i + 2] + + ptr[2 * i + 7] * filterCoeffs[i + 3]; + } + +#ifdef SOUNDTOUCH_INTEGER_SAMPLES + suml >>= resultDivFactor; + sumr >>= resultDivFactor; + // saturate to 16 bit integer limits + suml = (suml < -32768) ? -32768 : (suml > 32767) ? 32767 : suml; + // saturate to 16 bit integer limits + sumr = (sumr < -32768) ? -32768 : (sumr > 32767) ? 32767 : sumr; +#else + suml *= dScaler; + sumr *= dScaler; +#endif // SOUNDTOUCH_INTEGER_SAMPLES + dest[j] = (SAMPLETYPE)suml; + dest[j + 1] = (SAMPLETYPE)sumr; + } + return numSamples - length; +} + + + + +// Usual C-version of the filter routine for mono sound +uint FIRFilter::evaluateFilterMono(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples) const +{ + int j, end; +#ifdef SOUNDTOUCH_FLOAT_SAMPLES + // when using floating point samples, use a scaler instead of a divider + // because division is much slower operation than multiplying. + double dScaler = 1.0 / (double)resultDivider; +#endif + + assert(length != 0); + + end = numSamples - length; + #pragma omp parallel for + for (j = 0; j < end; j ++) + { + const SAMPLETYPE *pSrc = src + j; + LONG_SAMPLETYPE sum; + uint i; + + sum = 0; + for (i = 0; i < length; i += 4) + { + // loop is unrolled by factor of 4 here for efficiency + sum += pSrc[i + 0] * filterCoeffs[i + 0] + + pSrc[i + 1] * filterCoeffs[i + 1] + + pSrc[i + 2] * filterCoeffs[i + 2] + + pSrc[i + 3] * filterCoeffs[i + 3]; + } +#ifdef SOUNDTOUCH_INTEGER_SAMPLES + sum >>= resultDivFactor; + // saturate to 16 bit integer limits + sum = (sum < -32768) ? -32768 : (sum > 32767) ? 32767 : sum; +#else + sum *= dScaler; +#endif // SOUNDTOUCH_INTEGER_SAMPLES + dest[j] = (SAMPLETYPE)sum; + } + return end; +} + + +uint FIRFilter::evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples, uint numChannels) +{ + int j, end; + +#ifdef SOUNDTOUCH_FLOAT_SAMPLES + // when using floating point samples, use a scaler instead of a divider + // because division is much slower operation than multiplying. + double dScaler = 1.0 / (double)resultDivider; +#endif + + assert(length != 0); + assert(src != NULL); + assert(dest != NULL); + assert(filterCoeffs != NULL); + assert(numChannels < 16); + + end = numChannels * (numSamples - length); + + #pragma omp parallel for + for (j = 0; j < end; j += numChannels) + { + const SAMPLETYPE *ptr; + LONG_SAMPLETYPE sums[16]; + uint c, i; + + for (c = 0; c < numChannels; c ++) + { + sums[c] = 0; + } + + ptr = src + j; + + for (i = 0; i < length; i ++) + { + SAMPLETYPE coef=filterCoeffs[i]; + for (c = 0; c < numChannels; c ++) + { + sums[c] += ptr[0] * coef; + ptr ++; + } + } + + for (c = 0; c < numChannels; c ++) + { +#ifdef SOUNDTOUCH_INTEGER_SAMPLES + sums[c] >>= resultDivFactor; +#else + sums[c] *= dScaler; +#endif // SOUNDTOUCH_INTEGER_SAMPLES + dest[j+c] = (SAMPLETYPE)sums[c]; + } + } + return numSamples - length; +} + + +// Set filter coeffiecients and length. +// +// Throws an exception if filter length isn't divisible by 8 +void FIRFilter::setCoefficients(const SAMPLETYPE *coeffs, uint newLength, uint uResultDivFactor) +{ + assert(newLength > 0); + if (newLength % 8) ST_THROW_RT_ERROR("FIR filter length not divisible by 8"); + + lengthDiv8 = newLength / 8; + length = lengthDiv8 * 8; + assert(length == newLength); + + resultDivFactor = uResultDivFactor; + resultDivider = (SAMPLETYPE)::pow(2.0, (int)resultDivFactor); + + delete[] filterCoeffs; + filterCoeffs = new SAMPLETYPE[length]; + memcpy(filterCoeffs, coeffs, length * sizeof(SAMPLETYPE)); +} + + +uint FIRFilter::getLength() const +{ + return length; +} + + + +// Applies the filter to the given sequence of samples. +// +// Note : The amount of outputted samples is by value of 'filter_length' +// smaller than the amount of input samples. +uint FIRFilter::evaluate(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples, uint numChannels) +{ + assert(length > 0); + assert(lengthDiv8 * 8 == length); + + if (numSamples < length) return 0; + +#ifndef USE_MULTICH_ALWAYS + if (numChannels == 1) + { + return evaluateFilterMono(dest, src, numSamples); + } + else if (numChannels == 2) + { + return evaluateFilterStereo(dest, src, numSamples); + } + else +#endif // USE_MULTICH_ALWAYS + { + assert(numChannels > 0); + return evaluateFilterMulti(dest, src, numSamples, numChannels); + } +} + + + +// Operator 'new' is overloaded so that it automatically creates a suitable instance +// depending on if we've a MMX-capable CPU available or not. +void * FIRFilter::operator new(size_t s) +{ + // Notice! don't use "new FIRFilter" directly, use "newInstance" to create a new instance instead! + ST_THROW_RT_ERROR("Error in FIRFilter::new: Don't use 'new FIRFilter', use 'newInstance' member instead!"); + return newInstance(); +} + + +FIRFilter * FIRFilter::newInstance() +{ + uint uExtensions; + + uExtensions = detectCPUextensions(); + + // Check if MMX/SSE instruction set extensions supported by CPU + +#ifdef SOUNDTOUCH_ALLOW_MMX + // MMX routines available only with integer sample types + if (uExtensions & SUPPORT_MMX) + { + return ::new FIRFilterMMX; + } + else +#endif // SOUNDTOUCH_ALLOW_MMX + +#ifdef SOUNDTOUCH_ALLOW_SSE + if (uExtensions & SUPPORT_SSE) + { + // SSE support + return ::new FIRFilterSSE; + } + else +#endif // SOUNDTOUCH_ALLOW_SSE + + { + // ISA optimizations not supported, use plain C version + return ::new FIRFilter; + } +} diff --git a/src/FIRFilter.h b/src/FIRFilter.h new file mode 100644 index 0000000..158cdd2 --- /dev/null +++ b/src/FIRFilter.h @@ -0,0 +1,146 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// General FIR digital filter routines with MMX optimization. +/// +/// Note : MMX optimized functions reside in a separate, platform-specific file, +/// e.g. 'mmx_win.cpp' or 'mmx_gcc.cpp' +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2015-02-21 23:24:29 +0200 (Sat, 21 Feb 2015) $ +// File revision : $Revision: 4 $ +// +// $Id: FIRFilter.h 202 2015-02-21 21:24:29Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#ifndef FIRFilter_H +#define FIRFilter_H + +#include +#include "STTypes.h" + +namespace soundtouch +{ + +class FIRFilter +{ +protected: + // Number of FIR filter taps + uint length; + // Number of FIR filter taps divided by 8 + uint lengthDiv8; + + // Result divider factor in 2^k format + uint resultDivFactor; + + // Result divider value. + SAMPLETYPE resultDivider; + + // Memory for filter coefficients + SAMPLETYPE *filterCoeffs; + + virtual uint evaluateFilterStereo(SAMPLETYPE *dest, + const SAMPLETYPE *src, + uint numSamples) const; + virtual uint evaluateFilterMono(SAMPLETYPE *dest, + const SAMPLETYPE *src, + uint numSamples) const; + virtual uint evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples, uint numChannels); + +public: + FIRFilter(); + virtual ~FIRFilter(); + + /// Operator 'new' is overloaded so that it automatically creates a suitable instance + /// depending on if we've a MMX-capable CPU available or not. + static void * operator new(size_t s); + + static FIRFilter *newInstance(); + + /// Applies the filter to the given sequence of samples. + /// Note : The amount of outputted samples is by value of 'filter_length' + /// smaller than the amount of input samples. + /// + /// \return Number of samples copied to 'dest'. + uint evaluate(SAMPLETYPE *dest, + const SAMPLETYPE *src, + uint numSamples, + uint numChannels); + + uint getLength() const; + + virtual void setCoefficients(const SAMPLETYPE *coeffs, + uint newLength, + uint uResultDivFactor); +}; + + +// Optional subclasses that implement CPU-specific optimizations: + +#ifdef SOUNDTOUCH_ALLOW_MMX + +/// Class that implements MMX optimized functions exclusive for 16bit integer samples type. + class FIRFilterMMX : public FIRFilter + { + protected: + short *filterCoeffsUnalign; + short *filterCoeffsAlign; + + virtual uint evaluateFilterStereo(short *dest, const short *src, uint numSamples) const; + public: + FIRFilterMMX(); + ~FIRFilterMMX(); + + virtual void setCoefficients(const short *coeffs, uint newLength, uint uResultDivFactor); + }; + +#endif // SOUNDTOUCH_ALLOW_MMX + + +#ifdef SOUNDTOUCH_ALLOW_SSE + /// Class that implements SSE optimized functions exclusive for floating point samples type. + class FIRFilterSSE : public FIRFilter + { + protected: + float *filterCoeffsUnalign; + float *filterCoeffsAlign; + + virtual uint evaluateFilterStereo(float *dest, const float *src, uint numSamples) const; + public: + FIRFilterSSE(); + ~FIRFilterSSE(); + + virtual void setCoefficients(const float *coeffs, uint newLength, uint uResultDivFactor); + }; + +#endif // SOUNDTOUCH_ALLOW_SSE + +} + +#endif // FIRFilter_H diff --git a/src/InterpolateLinear.cpp b/src/InterpolateLinear.cpp new file mode 100644 index 0000000..8119813 --- /dev/null +++ b/src/InterpolateLinear.cpp @@ -0,0 +1,300 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// Linear interpolation algorithm. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// $Id: InterpolateLinear.cpp 225 2015-07-26 14:45:48Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#include +#include +#include "InterpolateLinear.h" + +using namespace soundtouch; + +////////////////////////////////////////////////////////////////////////////// +// +// InterpolateLinearInteger - integer arithmetic implementation +// + +/// fixed-point interpolation routine precision +#define SCALE 65536 + + +// Constructor +InterpolateLinearInteger::InterpolateLinearInteger() : TransposerBase() +{ + // Notice: use local function calling syntax for sake of clarity, + // to indicate the fact that C++ constructor can't call virtual functions. + resetRegisters(); + setRate(1.0f); +} + + +void InterpolateLinearInteger::resetRegisters() +{ + iFract = 0; +} + + +// Transposes the sample rate of the given samples using linear interpolation. +// 'Mono' version of the routine. Returns the number of samples returned in +// the "dest" buffer +int InterpolateLinearInteger::transposeMono(SAMPLETYPE *dest, const SAMPLETYPE *src, int &srcSamples) +{ + int i; + int srcSampleEnd = srcSamples - 1; + int srcCount = 0; + + i = 0; + while (srcCount < srcSampleEnd) + { + LONG_SAMPLETYPE temp; + + assert(iFract < SCALE); + + temp = (SCALE - iFract) * src[0] + iFract * src[1]; + dest[i] = (SAMPLETYPE)(temp / SCALE); + i++; + + iFract += iRate; + + int iWhole = iFract / SCALE; + iFract -= iWhole * SCALE; + srcCount += iWhole; + src += iWhole; + } + srcSamples = srcCount; + + return i; +} + + +// Transposes the sample rate of the given samples using linear interpolation. +// 'Stereo' version of the routine. Returns the number of samples returned in +// the "dest" buffer +int InterpolateLinearInteger::transposeStereo(SAMPLETYPE *dest, const SAMPLETYPE *src, int &srcSamples) +{ + int i; + int srcSampleEnd = srcSamples - 1; + int srcCount = 0; + + i = 0; + while (srcCount < srcSampleEnd) + { + LONG_SAMPLETYPE temp0; + LONG_SAMPLETYPE temp1; + + assert(iFract < SCALE); + + temp0 = (SCALE - iFract) * src[0] + iFract * src[2]; + temp1 = (SCALE - iFract) * src[1] + iFract * src[3]; + dest[0] = (SAMPLETYPE)(temp0 / SCALE); + dest[1] = (SAMPLETYPE)(temp1 / SCALE); + dest += 2; + i++; + + iFract += iRate; + + int iWhole = iFract / SCALE; + iFract -= iWhole * SCALE; + srcCount += iWhole; + src += 2*iWhole; + } + srcSamples = srcCount; + + return i; +} + + +int InterpolateLinearInteger::transposeMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, int &srcSamples) +{ + int i; + int srcSampleEnd = srcSamples - 1; + int srcCount = 0; + + i = 0; + while (srcCount < srcSampleEnd) + { + LONG_SAMPLETYPE temp, vol1; + + assert(iFract < SCALE); + vol1 = (SCALE - iFract); + for (int c = 0; c < numChannels; c ++) + { + temp = vol1 * src[c] + iFract * src[c + numChannels]; + dest[0] = (SAMPLETYPE)(temp / SCALE); + dest ++; + } + i++; + + iFract += iRate; + + int iWhole = iFract / SCALE; + iFract -= iWhole * SCALE; + srcCount += iWhole; + src += iWhole * numChannels; + } + srcSamples = srcCount; + + return i; +} + + +// Sets new target iRate. Normal iRate = 1.0, smaller values represent slower +// iRate, larger faster iRates. +void InterpolateLinearInteger::setRate(double newRate) +{ + iRate = (int)(newRate * SCALE + 0.5); + TransposerBase::setRate(newRate); +} + + +////////////////////////////////////////////////////////////////////////////// +// +// InterpolateLinearFloat - floating point arithmetic implementation +// +////////////////////////////////////////////////////////////////////////////// + + +// Constructor +InterpolateLinearFloat::InterpolateLinearFloat() : TransposerBase() +{ + // Notice: use local function calling syntax for sake of clarity, + // to indicate the fact that C++ constructor can't call virtual functions. + resetRegisters(); + setRate(1.0); +} + + +void InterpolateLinearFloat::resetRegisters() +{ + fract = 0; +} + + +// Transposes the sample rate of the given samples using linear interpolation. +// 'Mono' version of the routine. Returns the number of samples returned in +// the "dest" buffer +int InterpolateLinearFloat::transposeMono(SAMPLETYPE *dest, const SAMPLETYPE *src, int &srcSamples) +{ + int i; + int srcSampleEnd = srcSamples - 1; + int srcCount = 0; + + i = 0; + while (srcCount < srcSampleEnd) + { + double out; + assert(fract < 1.0); + + out = (1.0 - fract) * src[0] + fract * src[1]; + dest[i] = (SAMPLETYPE)out; + i ++; + + // update position fraction + fract += rate; + // update whole positions + int whole = (int)fract; + fract -= whole; + src += whole; + srcCount += whole; + } + srcSamples = srcCount; + return i; +} + + +// Transposes the sample rate of the given samples using linear interpolation. +// 'Mono' version of the routine. Returns the number of samples returned in +// the "dest" buffer +int InterpolateLinearFloat::transposeStereo(SAMPLETYPE *dest, const SAMPLETYPE *src, int &srcSamples) +{ + int i; + int srcSampleEnd = srcSamples - 1; + int srcCount = 0; + + i = 0; + while (srcCount < srcSampleEnd) + { + double out0, out1; + assert(fract < 1.0); + + out0 = (1.0 - fract) * src[0] + fract * src[2]; + out1 = (1.0 - fract) * src[1] + fract * src[3]; + dest[2*i] = (SAMPLETYPE)out0; + dest[2*i+1] = (SAMPLETYPE)out1; + i ++; + + // update position fraction + fract += rate; + // update whole positions + int whole = (int)fract; + fract -= whole; + src += 2*whole; + srcCount += whole; + } + srcSamples = srcCount; + return i; +} + + +int InterpolateLinearFloat::transposeMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, int &srcSamples) +{ + int i; + int srcSampleEnd = srcSamples - 1; + int srcCount = 0; + + i = 0; + while (srcCount < srcSampleEnd) + { + float temp, vol1, fract_float; + + vol1 = (float)(1.0 - fract); + fract_float = (float)fract; + for (int c = 0; c < numChannels; c ++) + { + temp = vol1 * src[c] + fract_float * src[c + numChannels]; + *dest = (SAMPLETYPE)temp; + dest ++; + } + i++; + + fract += rate; + + int iWhole = (int)fract; + fract -= iWhole; + srcCount += iWhole; + src += iWhole * numChannels; + } + srcSamples = srcCount; + + return i; +} diff --git a/src/InterpolateLinear.h b/src/InterpolateLinear.h new file mode 100644 index 0000000..6a7e11d --- /dev/null +++ b/src/InterpolateLinear.h @@ -0,0 +1,92 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// Linear interpolation routine. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// $Id: InterpolateLinear.h 225 2015-07-26 14:45:48Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#ifndef _InterpolateLinear_H_ +#define _InterpolateLinear_H_ + +#include "RateTransposer.h" +#include "STTypes.h" + +namespace soundtouch +{ + +/// Linear transposer class that uses integer arithmetics +class InterpolateLinearInteger : public TransposerBase +{ +protected: + int iFract; + int iRate; + + virtual void resetRegisters(); + + virtual int transposeMono(SAMPLETYPE *dest, + const SAMPLETYPE *src, + int &srcSamples); + virtual int transposeStereo(SAMPLETYPE *dest, + const SAMPLETYPE *src, + int &srcSamples); + virtual int transposeMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, int &srcSamples); +public: + InterpolateLinearInteger(); + + /// Sets new target rate. Normal rate = 1.0, smaller values represent slower + /// rate, larger faster rates. + virtual void setRate(double newRate); +}; + + +/// Linear transposer class that uses floating point arithmetics +class InterpolateLinearFloat : public TransposerBase +{ +protected: + double fract; + + virtual void resetRegisters(); + + virtual int transposeMono(SAMPLETYPE *dest, + const SAMPLETYPE *src, + int &srcSamples); + virtual int transposeStereo(SAMPLETYPE *dest, + const SAMPLETYPE *src, + int &srcSamples); + virtual int transposeMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, int &srcSamples); + +public: + InterpolateLinearFloat(); +}; + +} + +#endif diff --git a/src/RateTransposer.cpp b/src/RateTransposer.cpp new file mode 100644 index 0000000..6e0f03e --- /dev/null +++ b/src/RateTransposer.cpp @@ -0,0 +1,300 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// Sample rate transposer. Changes sample rate by using linear interpolation +/// together with anti-alias filtering (first order interpolation with anti- +/// alias filtering should be quite adequate for this application) +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2015-07-26 17:45:48 +0300 (Sun, 26 Jul 2015) $ +// File revision : $Revision: 4 $ +// +// $Id: RateTransposer.cpp 225 2015-07-26 14:45:48Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#include +#include +#include +#include +#include "RateTransposer.h" +#include "InterpolateLinear.h" +#include "AAFilter.h" + +using namespace soundtouch; + +// Define default interpolation algorithm here +TransposerBase::ALGORITHM TransposerBase::algorithm = TransposerBase::CUBIC; + + +// Constructor +RateTransposer::RateTransposer() : FIFOProcessor(&outputBuffer) +{ + bUseAAFilter = true; + + // Instantiates the anti-alias filter + pAAFilter = new AAFilter(64); + pTransposer = TransposerBase::newInstance(); +} + + + +RateTransposer::~RateTransposer() +{ + delete pAAFilter; + delete pTransposer; +} + + + +/// Enables/disables the anti-alias filter. Zero to disable, nonzero to enable +void RateTransposer::enableAAFilter(bool newMode) +{ + bUseAAFilter = newMode; +} + + +/// Returns nonzero if anti-alias filter is enabled. +bool RateTransposer::isAAFilterEnabled() const +{ + return bUseAAFilter; +} + + +AAFilter *RateTransposer::getAAFilter() +{ + return pAAFilter; +} + + + +// Sets new target iRate. Normal iRate = 1.0, smaller values represent slower +// iRate, larger faster iRates. +void RateTransposer::setRate(double newRate) +{ + double fCutoff; + + pTransposer->setRate(newRate); + + // design a new anti-alias filter + if (newRate > 1.0) + { + fCutoff = 0.5 / newRate; + } + else + { + fCutoff = 0.5 * newRate; + } + pAAFilter->setCutoffFreq(fCutoff); +} + + +// Adds 'nSamples' pcs of samples from the 'samples' memory position into +// the input of the object. +void RateTransposer::putSamples(const SAMPLETYPE *samples, uint nSamples) +{ + processSamples(samples, nSamples); +} + + +// Transposes sample rate by applying anti-alias filter to prevent folding. +// Returns amount of samples returned in the "dest" buffer. +// The maximum amount of samples that can be returned at a time is set by +// the 'set_returnBuffer_size' function. +void RateTransposer::processSamples(const SAMPLETYPE *src, uint nSamples) +{ + uint count; + + if (nSamples == 0) return; + + // Store samples to input buffer + inputBuffer.putSamples(src, nSamples); + + // If anti-alias filter is turned off, simply transpose without applying + // the filter + if (bUseAAFilter == false) + { + count = pTransposer->transpose(outputBuffer, inputBuffer); + return; + } + + assert(pAAFilter); + + // Transpose with anti-alias filter + if (pTransposer->rate < 1.0f) + { + // If the parameter 'Rate' value is smaller than 1, first transpose + // the samples and then apply the anti-alias filter to remove aliasing. + + // Transpose the samples, store the result to end of "midBuffer" + pTransposer->transpose(midBuffer, inputBuffer); + + // Apply the anti-alias filter for transposed samples in midBuffer + pAAFilter->evaluate(outputBuffer, midBuffer); + } + else + { + // If the parameter 'Rate' value is larger than 1, first apply the + // anti-alias filter to remove high frequencies (prevent them from folding + // over the lover frequencies), then transpose. + + // Apply the anti-alias filter for samples in inputBuffer + pAAFilter->evaluate(midBuffer, inputBuffer); + + // Transpose the AA-filtered samples in "midBuffer" + pTransposer->transpose(outputBuffer, midBuffer); + } +} + + +// Sets the number of channels, 1 = mono, 2 = stereo +void RateTransposer::setChannels(int nChannels) +{ + assert(nChannels > 0); + + if (pTransposer->numChannels == nChannels) return; + pTransposer->setChannels(nChannels); + + inputBuffer.setChannels(nChannels); + midBuffer.setChannels(nChannels); + outputBuffer.setChannels(nChannels); +} + + +// Clears all the samples in the object +void RateTransposer::clear() +{ + outputBuffer.clear(); + midBuffer.clear(); + inputBuffer.clear(); +} + + +// Returns nonzero if there aren't any samples available for outputting. +int RateTransposer::isEmpty() const +{ + int res; + + res = FIFOProcessor::isEmpty(); + if (res == 0) return 0; + return inputBuffer.isEmpty(); +} + + +////////////////////////////////////////////////////////////////////////////// +// +// TransposerBase - Base class for interpolation +// + +// static function to set interpolation algorithm +void TransposerBase::setAlgorithm(TransposerBase::ALGORITHM a) +{ + TransposerBase::algorithm = a; +} + + +// Transposes the sample rate of the given samples using linear interpolation. +// Returns the number of samples returned in the "dest" buffer +int TransposerBase::transpose(FIFOSampleBuffer &dest, FIFOSampleBuffer &src) +{ + int numSrcSamples = src.numSamples(); + int sizeDemand = (int)((double)numSrcSamples / rate) + 8; + int numOutput; + SAMPLETYPE *psrc = src.ptrBegin(); + SAMPLETYPE *pdest = dest.ptrEnd(sizeDemand); + +#ifndef USE_MULTICH_ALWAYS + if (numChannels == 1) + { + numOutput = transposeMono(pdest, psrc, numSrcSamples); + } + else if (numChannels == 2) + { + numOutput = transposeStereo(pdest, psrc, numSrcSamples); + } + else +#endif // USE_MULTICH_ALWAYS + { + assert(numChannels > 0); + numOutput = transposeMulti(pdest, psrc, numSrcSamples); + } + dest.putSamples(numOutput); + src.receiveSamples(numSrcSamples); + return numOutput; +} + + +TransposerBase::TransposerBase() +{ + numChannels = 0; + rate = 1.0f; +} + + +TransposerBase::~TransposerBase() +{ +} + + +void TransposerBase::setChannels(int channels) +{ + numChannels = channels; + resetRegisters(); +} + + +void TransposerBase::setRate(double newRate) +{ + rate = newRate; +} + + +// static factory function +TransposerBase *TransposerBase::newInstance() +{ +#ifdef SOUNDTOUCH_INTEGER_SAMPLES + // Notice: For integer arithmetics support only linear algorithm (due to simplest calculus) + return ::new InterpolateLinearInteger; +#else + switch (algorithm) + { + case LINEAR: + return new InterpolateLinearFloat; + + case CUBIC: + return new InterpolateCubic; + + case SHANNON: + return new InterpolateShannon; + + default: + assert(false); + return NULL; + } +#endif +} diff --git a/src/RateTransposer.h b/src/RateTransposer.h new file mode 100644 index 0000000..8d0eab8 --- /dev/null +++ b/src/RateTransposer.h @@ -0,0 +1,179 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// Sample rate transposer. Changes sample rate by using linear interpolation +/// together with anti-alias filtering (first order interpolation with anti- +/// alias filtering should be quite adequate for this application). +/// +/// Use either of the derived classes of 'RateTransposerInteger' or +/// 'RateTransposerFloat' for corresponding integer/floating point tranposing +/// algorithm implementation. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2015-07-26 17:45:48 +0300 (Sun, 26 Jul 2015) $ +// File revision : $Revision: 4 $ +// +// $Id: RateTransposer.h 225 2015-07-26 14:45:48Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#ifndef RateTransposer_H +#define RateTransposer_H + +#include +#include "AAFilter.h" +#include "FIFOSamplePipe.h" +#include "FIFOSampleBuffer.h" + +#include "STTypes.h" + +namespace soundtouch +{ + +/// Abstract base class for transposer implementations (linear, advanced vs integer, float etc) +class TransposerBase +{ +public: + enum ALGORITHM { + LINEAR = 0, + CUBIC, + SHANNON + }; + +protected: + virtual void resetRegisters() = 0; + + virtual int transposeMono(SAMPLETYPE *dest, + const SAMPLETYPE *src, + int &srcSamples) = 0; + virtual int transposeStereo(SAMPLETYPE *dest, + const SAMPLETYPE *src, + int &srcSamples) = 0; + virtual int transposeMulti(SAMPLETYPE *dest, + const SAMPLETYPE *src, + int &srcSamples) = 0; + + static ALGORITHM algorithm; + +public: + double rate; + int numChannels; + + TransposerBase(); + virtual ~TransposerBase(); + + virtual int transpose(FIFOSampleBuffer &dest, FIFOSampleBuffer &src); + virtual void setRate(double newRate); + virtual void setChannels(int channels); + + // static factory function + static TransposerBase *newInstance(); + + // static function to set interpolation algorithm + static void setAlgorithm(ALGORITHM a); +}; + + +/// A common linear samplerate transposer class. +/// +class RateTransposer : public FIFOProcessor +{ +protected: + /// Anti-alias filter object + AAFilter *pAAFilter; + TransposerBase *pTransposer; + + /// Buffer for collecting samples to feed the anti-alias filter between + /// two batches + FIFOSampleBuffer inputBuffer; + + /// Buffer for keeping samples between transposing & anti-alias filter + FIFOSampleBuffer midBuffer; + + /// Output sample buffer + FIFOSampleBuffer outputBuffer; + + bool bUseAAFilter; + + + /// Transposes sample rate by applying anti-alias filter to prevent folding. + /// Returns amount of samples returned in the "dest" buffer. + /// The maximum amount of samples that can be returned at a time is set by + /// the 'set_returnBuffer_size' function. + void processSamples(const SAMPLETYPE *src, + uint numSamples); + +public: + RateTransposer(); + virtual ~RateTransposer(); + + /// Operator 'new' is overloaded so that it automatically creates a suitable instance + /// depending on if we're to use integer or floating point arithmetics. +// static void *operator new(size_t s); + + /// Use this function instead of "new" operator to create a new instance of this class. + /// This function automatically chooses a correct implementation, depending on if + /// integer ot floating point arithmetics are to be used. +// static RateTransposer *newInstance(); + + /// Returns the output buffer object + FIFOSamplePipe *getOutput() { return &outputBuffer; }; + + /// Returns the store buffer object +// FIFOSamplePipe *getStore() { return &storeBuffer; }; + + /// Return anti-alias filter object + AAFilter *getAAFilter(); + + /// Enables/disables the anti-alias filter. Zero to disable, nonzero to enable + void enableAAFilter(bool newMode); + + /// Returns nonzero if anti-alias filter is enabled. + bool isAAFilterEnabled() const; + + /// Sets new target rate. Normal rate = 1.0, smaller values represent slower + /// rate, larger faster rates. + virtual void setRate(double newRate); + + /// Sets the number of channels, 1 = mono, 2 = stereo + void setChannels(int channels); + + /// Adds 'numSamples' pcs of samples from the 'samples' memory position into + /// the input of the object. + void putSamples(const SAMPLETYPE *samples, uint numSamples); + + /// Clears all the samples in the object + void clear(); + + /// Returns nonzero if there aren't any samples available for outputting. + int isEmpty() const; +}; + +} + +#endif diff --git a/src/SoundTouch.cpp b/src/SoundTouch.cpp new file mode 100644 index 0000000..2c715e2 --- /dev/null +++ b/src/SoundTouch.cpp @@ -0,0 +1,526 @@ +////////////////////////////////////////////////////////////////////////////// +/// +/// SoundTouch - main class for tempo/pitch/rate adjusting routines. +/// +/// Notes: +/// - Initialize the SoundTouch object instance by setting up the sound stream +/// parameters with functions 'setSampleRate' and 'setChannels', then set +/// desired tempo/pitch/rate settings with the corresponding functions. +/// +/// - The SoundTouch class behaves like a first-in-first-out pipeline: The +/// samples that are to be processed are fed into one of the pipe by calling +/// function 'putSamples', while the ready processed samples can be read +/// from the other end of the pipeline with function 'receiveSamples'. +/// +/// - The SoundTouch processing classes require certain sized 'batches' of +/// samples in order to process the sound. For this reason the classes buffer +/// incoming samples until there are enough of samples available for +/// processing, then they carry out the processing step and consequently +/// make the processed samples available for outputting. +/// +/// - For the above reason, the processing routines introduce a certain +/// 'latency' between the input and output, so that the samples input to +/// SoundTouch may not be immediately available in the output, and neither +/// the amount of outputtable samples may not immediately be in direct +/// relationship with the amount of previously input samples. +/// +/// - The tempo/pitch/rate control parameters can be altered during processing. +/// Please notice though that they aren't currently protected by semaphores, +/// so in multi-thread application external semaphore protection may be +/// required. +/// +/// - This class utilizes classes 'TDStretch' for tempo change (without modifying +/// pitch) and 'RateTransposer' for changing the playback rate (that is, both +/// tempo and pitch in the same ratio) of the sound. The third available control +/// 'pitch' (change pitch but maintain tempo) is produced by a combination of +/// combining the two other controls. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2015-07-26 17:45:48 +0300 (Sun, 26 Jul 2015) $ +// File revision : $Revision: 4 $ +// +// $Id: SoundTouch.cpp 225 2015-07-26 14:45:48Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#include +#include +#include +#include +#include + +#include "SoundTouch.h" +#include "TDStretch.h" +#include "RateTransposer.h" +#include "cpu_detect.h" + +using namespace soundtouch; + +/// test if two floating point numbers are equal +#define TEST_FLOAT_EQUAL(a, b) (fabs(a - b) < 1e-10) + + +/// Print library version string for autoconf +extern "C" void soundtouch_ac_test() +{ + printf("SoundTouch Version: %s\n",SOUNDTOUCH_VERSION); +} + + +SoundTouch::SoundTouch() +{ + // Initialize rate transposer and tempo changer instances + + pRateTransposer = new RateTransposer(); + pTDStretch = TDStretch::newInstance(); + + setOutPipe(pTDStretch); + + rate = tempo = 0; + + virtualPitch = + virtualRate = + virtualTempo = 1.0; + + calcEffectiveRateAndTempo(); + + samplesExpectedOut = 0; + samplesOutput = 0; + + channels = 0; + bSrateSet = false; +} + + + +SoundTouch::~SoundTouch() +{ + delete pRateTransposer; + delete pTDStretch; +} + + + +/// Get SoundTouch library version string +const char *SoundTouch::getVersionString() +{ + static const char *_version = SOUNDTOUCH_VERSION; + + return _version; +} + + +/// Get SoundTouch library version Id +uint SoundTouch::getVersionId() +{ + return SOUNDTOUCH_VERSION_ID; +} + + +// Sets the number of channels, 1 = mono, 2 = stereo +void SoundTouch::setChannels(uint numChannels) +{ + /*if (numChannels != 1 && numChannels != 2) + { + //ST_THROW_RT_ERROR("Illegal number of channels"); + return; + }*/ + channels = numChannels; + pRateTransposer->setChannels((int)numChannels); + pTDStretch->setChannels((int)numChannels); +} + + + +// Sets new rate control value. Normal rate = 1.0, smaller values +// represent slower rate, larger faster rates. +void SoundTouch::setRate(double newRate) +{ + virtualRate = newRate; + calcEffectiveRateAndTempo(); +} + + + +// Sets new rate control value as a difference in percents compared +// to the original rate (-50 .. +100 %) +void SoundTouch::setRateChange(double newRate) +{ + virtualRate = 1.0 + 0.01 * newRate; + calcEffectiveRateAndTempo(); +} + + + +// Sets new tempo control value. Normal tempo = 1.0, smaller values +// represent slower tempo, larger faster tempo. +void SoundTouch::setTempo(double newTempo) +{ + virtualTempo = newTempo; + calcEffectiveRateAndTempo(); +} + + + +// Sets new tempo control value as a difference in percents compared +// to the original tempo (-50 .. +100 %) +void SoundTouch::setTempoChange(double newTempo) +{ + virtualTempo = 1.0 + 0.01 * newTempo; + calcEffectiveRateAndTempo(); +} + + + +// Sets new pitch control value. Original pitch = 1.0, smaller values +// represent lower pitches, larger values higher pitch. +void SoundTouch::setPitch(double newPitch) +{ + virtualPitch = newPitch; + calcEffectiveRateAndTempo(); +} + + + +// Sets pitch change in octaves compared to the original pitch +// (-1.00 .. +1.00) +void SoundTouch::setPitchOctaves(double newPitch) +{ + virtualPitch = exp(0.69314718056 * newPitch); + calcEffectiveRateAndTempo(); +} + + + +// Sets pitch change in semi-tones compared to the original pitch +// (-12 .. +12) +void SoundTouch::setPitchSemiTones(int newPitch) +{ + setPitchOctaves((double)newPitch / 12.0); +} + + + +void SoundTouch::setPitchSemiTones(double newPitch) +{ + setPitchOctaves(newPitch / 12.0); +} + + +// Calculates 'effective' rate and tempo values from the +// nominal control values. +void SoundTouch::calcEffectiveRateAndTempo() +{ + double oldTempo = tempo; + double oldRate = rate; + + tempo = virtualTempo / virtualPitch; + rate = virtualPitch * virtualRate; + + if (!TEST_FLOAT_EQUAL(rate,oldRate)) pRateTransposer->setRate(rate); + if (!TEST_FLOAT_EQUAL(tempo, oldTempo)) pTDStretch->setTempo(tempo); + +#ifndef SOUNDTOUCH_PREVENT_CLICK_AT_RATE_CROSSOVER + if (rate <= 1.0f) + { + if (output != pTDStretch) + { + FIFOSamplePipe *tempoOut; + + assert(output == pRateTransposer); + // move samples in the current output buffer to the output of pTDStretch + tempoOut = pTDStretch->getOutput(); + tempoOut->moveSamples(*output); + // move samples in pitch transposer's store buffer to tempo changer's input + // deprecated : pTDStretch->moveSamples(*pRateTransposer->getStore()); + + output = pTDStretch; + } + } + else +#endif + { + if (output != pRateTransposer) + { + FIFOSamplePipe *transOut; + + assert(output == pTDStretch); + // move samples in the current output buffer to the output of pRateTransposer + transOut = pRateTransposer->getOutput(); + transOut->moveSamples(*output); + // move samples in tempo changer's input to pitch transposer's input + pRateTransposer->moveSamples(*pTDStretch->getInput()); + + output = pRateTransposer; + } + } +} + + +// Sets sample rate. +void SoundTouch::setSampleRate(uint srate) +{ + bSrateSet = true; + // set sample rate, leave other tempo changer parameters as they are. + pTDStretch->setParameters((int)srate); +} + + +// Adds 'numSamples' pcs of samples from the 'samples' memory position into +// the input of the object. +void SoundTouch::putSamples(const SAMPLETYPE *samples, uint nSamples) +{ + if (bSrateSet == false) + { + ST_THROW_RT_ERROR("SoundTouch : Sample rate not defined"); + } + else if (channels == 0) + { + ST_THROW_RT_ERROR("SoundTouch : Number of channels not defined"); + } + + // Transpose the rate of the new samples if necessary + /* Bypass the nominal setting - can introduce a click in sound when tempo/pitch control crosses the nominal value... + if (rate == 1.0f) + { + // The rate value is same as the original, simply evaluate the tempo changer. + assert(output == pTDStretch); + if (pRateTransposer->isEmpty() == 0) + { + // yet flush the last samples in the pitch transposer buffer + // (may happen if 'rate' changes from a non-zero value to zero) + pTDStretch->moveSamples(*pRateTransposer); + } + pTDStretch->putSamples(samples, nSamples); + } + */ + + // accumulate how many samples are expected out from processing, given the current + // processing setting + samplesExpectedOut += (double)nSamples / ((double)rate * (double)tempo); + +#ifndef SOUNDTOUCH_PREVENT_CLICK_AT_RATE_CROSSOVER + if (rate <= 1.0f) + { + // transpose the rate down, output the transposed sound to tempo changer buffer + assert(output == pTDStretch); + pRateTransposer->putSamples(samples, nSamples); + pTDStretch->moveSamples(*pRateTransposer); + } + else +#endif + { + // evaluate the tempo changer, then transpose the rate up, + assert(output == pRateTransposer); + pTDStretch->putSamples(samples, nSamples); + pRateTransposer->moveSamples(*pTDStretch); + } +} + + +// Flushes the last samples from the processing pipeline to the output. +// Clears also the internal processing buffers. +// +// Note: This function is meant for extracting the last samples of a sound +// stream. This function may introduce additional blank samples in the end +// of the sound stream, and thus it's not recommended to call this function +// in the middle of a sound stream. +void SoundTouch::flush() +{ + int i; + int numStillExpected; + SAMPLETYPE *buff = new SAMPLETYPE[128 * channels]; + + // how many samples are still expected to output + numStillExpected = (int)((long)(samplesExpectedOut + 0.5) - samplesOutput); + + memset(buff, 0, 128 * channels * sizeof(SAMPLETYPE)); + // "Push" the last active samples out from the processing pipeline by + // feeding blank samples into the processing pipeline until new, + // processed samples appear in the output (not however, more than + // 24ksamples in any case) + for (i = 0; (numStillExpected > (int)numSamples()) && (i < 200); i ++) + { + putSamples(buff, 128); + } + + adjustAmountOfSamples(numStillExpected); + + delete[] buff; + + // Clear input buffers + // pRateTransposer->clearInput(); + pTDStretch->clearInput(); + // yet leave the output intouched as that's where the + // flushed samples are! +} + + +// Changes a setting controlling the processing system behaviour. See the +// 'SETTING_...' defines for available setting ID's. +bool SoundTouch::setSetting(int settingId, int value) +{ + int sampleRate, sequenceMs, seekWindowMs, overlapMs; + + // read current tdstretch routine parameters + pTDStretch->getParameters(&sampleRate, &sequenceMs, &seekWindowMs, &overlapMs); + + switch (settingId) + { + case SETTING_USE_AA_FILTER : + // enables / disabless anti-alias filter + pRateTransposer->enableAAFilter((value != 0) ? true : false); + return true; + + case SETTING_AA_FILTER_LENGTH : + // sets anti-alias filter length + pRateTransposer->getAAFilter()->setLength(value); + return true; + + case SETTING_USE_QUICKSEEK : + // enables / disables tempo routine quick seeking algorithm + pTDStretch->enableQuickSeek((value != 0) ? true : false); + return true; + + case SETTING_SEQUENCE_MS: + // change time-stretch sequence duration parameter + pTDStretch->setParameters(sampleRate, value, seekWindowMs, overlapMs); + return true; + + case SETTING_SEEKWINDOW_MS: + // change time-stretch seek window length parameter + pTDStretch->setParameters(sampleRate, sequenceMs, value, overlapMs); + return true; + + case SETTING_OVERLAP_MS: + // change time-stretch overlap length parameter + pTDStretch->setParameters(sampleRate, sequenceMs, seekWindowMs, value); + return true; + + default : + return false; + } +} + + +// Reads a setting controlling the processing system behaviour. See the +// 'SETTING_...' defines for available setting ID's. +// +// Returns the setting value. +int SoundTouch::getSetting(int settingId) const +{ + int temp; + + switch (settingId) + { + case SETTING_USE_AA_FILTER : + return (uint)pRateTransposer->isAAFilterEnabled(); + + case SETTING_AA_FILTER_LENGTH : + return pRateTransposer->getAAFilter()->getLength(); + + case SETTING_USE_QUICKSEEK : + return (uint) pTDStretch->isQuickSeekEnabled(); + + case SETTING_SEQUENCE_MS: + pTDStretch->getParameters(NULL, &temp, NULL, NULL); + return temp; + + case SETTING_SEEKWINDOW_MS: + pTDStretch->getParameters(NULL, NULL, &temp, NULL); + return temp; + + case SETTING_OVERLAP_MS: + pTDStretch->getParameters(NULL, NULL, NULL, &temp); + return temp; + + case SETTING_NOMINAL_INPUT_SEQUENCE : + return pTDStretch->getInputSampleReq(); + + case SETTING_NOMINAL_OUTPUT_SEQUENCE : + return pTDStretch->getOutputBatchSize(); + + default : + return 0; + } +} + + +// Clears all the samples in the object's output and internal processing +// buffers. +void SoundTouch::clear() +{ + samplesExpectedOut = 0; + pRateTransposer->clear(); + pTDStretch->clear(); +} + + + +/// Returns number of samples currently unprocessed. +uint SoundTouch::numUnprocessedSamples() const +{ + FIFOSamplePipe * psp; + if (pTDStretch) + { + psp = pTDStretch->getInput(); + if (psp) + { + return psp->numSamples(); + } + } + return 0; +} + + + +/// Output samples from beginning of the sample buffer. Copies requested samples to +/// output buffer and removes them from the sample buffer. If there are less than +/// 'numsample' samples in the buffer, returns all that available. +/// +/// \return Number of samples returned. +uint SoundTouch::receiveSamples(SAMPLETYPE *output, uint maxSamples) +{ + uint ret = FIFOProcessor::receiveSamples(output, maxSamples); + samplesOutput += (long)ret; + return ret; +} + + +/// Adjusts book-keeping so that given number of samples are removed from beginning of the +/// sample buffer without copying them anywhere. +/// +/// Used to reduce the number of samples in the buffer when accessing the sample buffer directly +/// with 'ptrBegin' function. +uint SoundTouch::receiveSamples(uint maxSamples) +{ + uint ret = FIFOProcessor::receiveSamples(maxSamples); + samplesOutput += (long)ret; + return ret; +} diff --git a/src/TDStretch.cpp b/src/TDStretch.cpp new file mode 100644 index 0000000..d030558 --- /dev/null +++ b/src/TDStretch.cpp @@ -0,0 +1,1078 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// Sampled sound tempo changer/time stretch algorithm. Changes the sound tempo +/// while maintaining the original pitch by using a time domain WSOLA-like +/// method with several performance-increasing tweaks. +/// +/// Note : MMX optimized functions reside in a separate, platform-specific +/// file, e.g. 'mmx_win.cpp' or 'mmx_gcc.cpp' +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2015-08-09 00:00:15 +0300 (Sun, 09 Aug 2015) $ +// File revision : $Revision: 1.12 $ +// +// $Id: TDStretch.cpp 226 2015-08-08 21:00:15Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#include +#include +#include +#include +#include + +#include "STTypes.h" +#include "cpu_detect.h" +#include "TDStretch.h" + +using namespace soundtouch; + +#define max(x, y) (((x) > (y)) ? (x) : (y)) + + +/***************************************************************************** + * + * Constant definitions + * + *****************************************************************************/ + +// Table for the hierarchical mixing position seeking algorithm +const short _scanOffsets[5][24]={ + { 124, 186, 248, 310, 372, 434, 496, 558, 620, 682, 744, 806, + 868, 930, 992, 1054, 1116, 1178, 1240, 1302, 1364, 1426, 1488, 0}, + {-100, -75, -50, -25, 25, 50, 75, 100, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + { -20, -15, -10, -5, 5, 10, 15, 20, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + { -4, -3, -2, -1, 1, 2, 3, 4, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + { 121, 114, 97, 114, 98, 105, 108, 32, 104, 99, 117, 111, + 116, 100, 110, 117, 111, 115, 0, 0, 0, 0, 0, 0}}; + +/***************************************************************************** + * + * Implementation of the class 'TDStretch' + * + *****************************************************************************/ + + +TDStretch::TDStretch() : FIFOProcessor(&outputBuffer) +{ + bQuickSeek = false; + channels = 2; + + pMidBuffer = NULL; + pMidBufferUnaligned = NULL; + overlapLength = 0; + + bAutoSeqSetting = true; + bAutoSeekSetting = true; + + maxnorm = 0; + maxnormf = 1e8; + + skipFract = 0; + + tempo = 1.0f; + setParameters(44100, DEFAULT_SEQUENCE_MS, DEFAULT_SEEKWINDOW_MS, DEFAULT_OVERLAP_MS); + setTempo(1.0f); + + clear(); +} + + + +TDStretch::~TDStretch() +{ + delete[] pMidBufferUnaligned; +} + + + +// Sets routine control parameters. These control are certain time constants +// defining how the sound is stretched to the desired duration. +// +// 'sampleRate' = sample rate of the sound +// 'sequenceMS' = one processing sequence length in milliseconds (default = 82 ms) +// 'seekwindowMS' = seeking window length for scanning the best overlapping +// position (default = 28 ms) +// 'overlapMS' = overlapping length (default = 12 ms) + +void TDStretch::setParameters(int aSampleRate, int aSequenceMS, + int aSeekWindowMS, int aOverlapMS) +{ + // accept only positive parameter values - if zero or negative, use old values instead + if (aSampleRate > 0) this->sampleRate = aSampleRate; + if (aOverlapMS > 0) this->overlapMs = aOverlapMS; + + if (aSequenceMS > 0) + { + this->sequenceMs = aSequenceMS; + bAutoSeqSetting = false; + } + else if (aSequenceMS == 0) + { + // if zero, use automatic setting + bAutoSeqSetting = true; + } + + if (aSeekWindowMS > 0) + { + this->seekWindowMs = aSeekWindowMS; + bAutoSeekSetting = false; + } + else if (aSeekWindowMS == 0) + { + // if zero, use automatic setting + bAutoSeekSetting = true; + } + + calcSeqParameters(); + + calculateOverlapLength(overlapMs); + + // set tempo to recalculate 'sampleReq' + setTempo(tempo); +} + + + +/// Get routine control parameters, see setParameters() function. +/// Any of the parameters to this function can be NULL, in such case corresponding parameter +/// value isn't returned. +void TDStretch::getParameters(int *pSampleRate, int *pSequenceMs, int *pSeekWindowMs, int *pOverlapMs) const +{ + if (pSampleRate) + { + *pSampleRate = sampleRate; + } + + if (pSequenceMs) + { + *pSequenceMs = (bAutoSeqSetting) ? (USE_AUTO_SEQUENCE_LEN) : sequenceMs; + } + + if (pSeekWindowMs) + { + *pSeekWindowMs = (bAutoSeekSetting) ? (USE_AUTO_SEEKWINDOW_LEN) : seekWindowMs; + } + + if (pOverlapMs) + { + *pOverlapMs = overlapMs; + } +} + + +// Overlaps samples in 'midBuffer' with the samples in 'pInput' +void TDStretch::overlapMono(SAMPLETYPE *pOutput, const SAMPLETYPE *pInput) const +{ + int i; + SAMPLETYPE m1, m2; + + m1 = (SAMPLETYPE)0; + m2 = (SAMPLETYPE)overlapLength; + + for (i = 0; i < overlapLength ; i ++) + { + pOutput[i] = (pInput[i] * m1 + pMidBuffer[i] * m2 ) / overlapLength; + m1 += 1; + m2 -= 1; + } +} + + + +void TDStretch::clearMidBuffer() +{ + memset(pMidBuffer, 0, channels * sizeof(SAMPLETYPE) * overlapLength); +} + + +void TDStretch::clearInput() +{ + inputBuffer.clear(); + clearMidBuffer(); +} + + +// Clears the sample buffers +void TDStretch::clear() +{ + outputBuffer.clear(); + clearInput(); +} + + + +// Enables/disables the quick position seeking algorithm. Zero to disable, nonzero +// to enable +void TDStretch::enableQuickSeek(bool enable) +{ + bQuickSeek = enable; +} + + +// Returns nonzero if the quick seeking algorithm is enabled. +bool TDStretch::isQuickSeekEnabled() const +{ + return bQuickSeek; +} + + +// Seeks for the optimal overlap-mixing position. +int TDStretch::seekBestOverlapPosition(const SAMPLETYPE *refPos) +{ + if (bQuickSeek) + { + return seekBestOverlapPositionQuick(refPos); + } + else + { + return seekBestOverlapPositionFull(refPos); + } +} + + +// Overlaps samples in 'midBuffer' with the samples in 'pInputBuffer' at position +// of 'ovlPos'. +inline void TDStretch::overlap(SAMPLETYPE *pOutput, const SAMPLETYPE *pInput, uint ovlPos) const +{ +#ifndef USE_MULTICH_ALWAYS + if (channels == 1) + { + // mono sound. + overlapMono(pOutput, pInput + ovlPos); + } + else if (channels == 2) + { + // stereo sound + overlapStereo(pOutput, pInput + 2 * ovlPos); + } + else +#endif // USE_MULTICH_ALWAYS + { + assert(channels > 0); + overlapMulti(pOutput, pInput + channels * ovlPos); + } +} + + +// Seeks for the optimal overlap-mixing position. The 'stereo' version of the +// routine +// +// The best position is determined as the position where the two overlapped +// sample sequences are 'most alike', in terms of the highest cross-correlation +// value over the overlapping period +int TDStretch::seekBestOverlapPositionFull(const SAMPLETYPE *refPos) +{ + int bestOffs; + double bestCorr; + int i; + double norm; + + bestCorr = FLT_MIN; + bestOffs = 0; + + // Scans for the best correlation value by testing each possible position + // over the permitted range. + bestCorr = calcCrossCorr(refPos, pMidBuffer, norm); + + #pragma omp parallel for + for (i = 1; i < seekLength; i ++) + { + double corr; + // Calculates correlation value for the mixing position corresponding to 'i' +#ifdef _OPENMP + // in parallel OpenMP mode, can't use norm accumulator version as parallel executor won't + // iterate the loop in sequential order + corr = calcCrossCorr(refPos + channels * i, pMidBuffer, norm); +#else + // In non-parallel version call "calcCrossCorrAccumulate" that is otherwise same + // as "calcCrossCorr", but saves time by reusing & updating previously stored + // "norm" value + corr = calcCrossCorrAccumulate(refPos + channels * i, pMidBuffer, norm); +#endif + // heuristic rule to slightly favour values close to mid of the range + double tmp = (double)(2 * i - seekLength) / (double)seekLength; + corr = ((corr + 0.1) * (1.0 - 0.25 * tmp * tmp)); + + // Checks for the highest correlation value + if (corr > bestCorr) + { + // For optimal performance, enter critical section only in case that best value found. + // in such case repeat 'if' condition as it's possible that parallel execution may have + // updated the bestCorr value in the mean time + #pragma omp critical + if (corr > bestCorr) + { + bestCorr = corr; + bestOffs = i; + } + } + } + +#ifdef SOUNDTOUCH_INTEGER_SAMPLES + adaptNormalizer(); +#endif + + // clear cross correlation routine state if necessary (is so e.g. in MMX routines). + clearCrossCorrState(); + + return bestOffs; +} + + +// Quick seek algorithm for improved runtime-performance: First roughly scans through the +// correlation area, and then scan surroundings of two best preliminary correlation candidates +// with improved precision +// +// Based on testing: +// - This algorithm gives on average 99% as good match as the full algorith +// - this quick seek algorithm finds the best match on ~90% of cases +// - on those 10% of cases when this algorithm doesn't find best match, +// it still finds on average ~90% match vs. the best possible match +int TDStretch::seekBestOverlapPositionQuick(const SAMPLETYPE *refPos) +{ +#define _MIN(a, b) (((a) < (b)) ? (a) : (b)) +#define SCANSTEP 16 +#define SCANWIND 8 + + int bestOffs; + int i; + int bestOffs2; + float bestCorr, corr; + float bestCorr2; + double norm; + + // note: 'float' types used in this function in case that the platform would need to use software-fp + + bestCorr = FLT_MIN; + bestOffs = SCANWIND; + bestCorr2 = FLT_MIN; + bestOffs2 = 0; + + int best = 0; + + // Scans for the best correlation value by testing each possible position + // over the permitted range. Look for two best matches on the first pass to + // increase possibility of ideal match. + // + // Begin from "SCANSTEP" instead of SCANWIND to make the calculation + // catch the 'middlepoint' of seekLength vector as that's the a-priori + // expected best match position + // + // Roughly: + // - 15% of cases find best result directly on the first round, + // - 75% cases find better match on 2nd round around the best match from 1st round + // - 10% cases find better match on 2nd round around the 2nd-best-match from 1st round + for (i = SCANSTEP; i < seekLength - SCANWIND - 1; i += SCANSTEP) + { + // Calculates correlation value for the mixing position corresponding + // to 'i' + corr = (float)calcCrossCorr(refPos + channels*i, pMidBuffer, norm); + // heuristic rule to slightly favour values close to mid of the seek range + float tmp = (float)(2 * i - seekLength - 1) / (float)seekLength; + corr = ((corr + 0.1f) * (1.0f - 0.25f * tmp * tmp)); + + // Checks for the highest correlation value + if (corr > bestCorr) + { + // found new best match. keep the previous best as 2nd best match + bestCorr2 = bestCorr; + bestOffs2 = bestOffs; + bestCorr = corr; + bestOffs = i; + } + else if (corr > bestCorr2) + { + // not new best, but still new 2nd best match + bestCorr2 = corr; + bestOffs2 = i; + } + } + + // Scans surroundings of the found best match with small stepping + int end = _MIN(bestOffs + SCANWIND + 1, seekLength); + for (i = bestOffs - SCANWIND; i < end; i++) + { + if (i == bestOffs) continue; // this offset already calculated, thus skip + + // Calculates correlation value for the mixing position corresponding + // to 'i' + corr = (float)calcCrossCorr(refPos + channels*i, pMidBuffer, norm); + // heuristic rule to slightly favour values close to mid of the range + float tmp = (float)(2 * i - seekLength - 1) / (float)seekLength; + corr = ((corr + 0.1f) * (1.0f - 0.25f * tmp * tmp)); + + // Checks for the highest correlation value + if (corr > bestCorr) + { + bestCorr = corr; + bestOffs = i; + best = 1; + } + } + + // Scans surroundings of the 2nd best match with small stepping + end = _MIN(bestOffs2 + SCANWIND + 1, seekLength); + for (i = bestOffs2 - SCANWIND; i < end; i++) + { + if (i == bestOffs2) continue; // this offset already calculated, thus skip + + // Calculates correlation value for the mixing position corresponding + // to 'i' + corr = (float)calcCrossCorr(refPos + channels*i, pMidBuffer, norm); + // heuristic rule to slightly favour values close to mid of the range + float tmp = (float)(2 * i - seekLength - 1) / (float)seekLength; + corr = ((corr + 0.1f) * (1.0f - 0.25f * tmp * tmp)); + + // Checks for the highest correlation value + if (corr > bestCorr) + { + bestCorr = corr; + bestOffs = i; + best = 2; + } + } + + // clear cross correlation routine state if necessary (is so e.g. in MMX routines). + clearCrossCorrState(); + +#ifdef SOUNDTOUCH_INTEGER_SAMPLES + adaptNormalizer(); +#endif + + return bestOffs; +} + + + + +/// For integer algorithm: adapt normalization factor divider with music so that +/// it'll not be pessimistically restrictive that can degrade quality on quieter sections +/// yet won't cause integer overflows either +void TDStretch::adaptNormalizer() +{ + // Do not adapt normalizer over too silent sequences to avoid averaging filter depleting to + // too low values during pauses in music + if ((maxnorm > 1000) || (maxnormf > 40000000)) + { + //norm averaging filter + maxnormf = 0.9f * maxnormf + 0.1f * (float)maxnorm; + + if ((maxnorm > 800000000) && (overlapDividerBitsNorm < 16)) + { + // large values, so increase divider + overlapDividerBitsNorm++; + if (maxnorm > 1600000000) overlapDividerBitsNorm++; // extra large value => extra increase + } + else if ((maxnormf < 1000000) && (overlapDividerBitsNorm > 0)) + { + // extra small values, decrease divider + overlapDividerBitsNorm--; + } + } + + maxnorm = 0; +} + + +/// clear cross correlation routine state if necessary +void TDStretch::clearCrossCorrState() +{ + // default implementation is empty. +} + + +/// Calculates processing sequence length according to tempo setting +void TDStretch::calcSeqParameters() +{ + // Adjust tempo param according to tempo, so that variating processing sequence length is used + // at varius tempo settings, between the given low...top limits + #define AUTOSEQ_TEMPO_LOW 0.5 // auto setting low tempo range (-50%) + #define AUTOSEQ_TEMPO_TOP 2.0 // auto setting top tempo range (+100%) + + // sequence-ms setting values at above low & top tempo + #define AUTOSEQ_AT_MIN 125.0 + #define AUTOSEQ_AT_MAX 50.0 + #define AUTOSEQ_K ((AUTOSEQ_AT_MAX - AUTOSEQ_AT_MIN) / (AUTOSEQ_TEMPO_TOP - AUTOSEQ_TEMPO_LOW)) + #define AUTOSEQ_C (AUTOSEQ_AT_MIN - (AUTOSEQ_K) * (AUTOSEQ_TEMPO_LOW)) + + // seek-window-ms setting values at above low & top tempoq + #define AUTOSEEK_AT_MIN 25.0 + #define AUTOSEEK_AT_MAX 15.0 + #define AUTOSEEK_K ((AUTOSEEK_AT_MAX - AUTOSEEK_AT_MIN) / (AUTOSEQ_TEMPO_TOP - AUTOSEQ_TEMPO_LOW)) + #define AUTOSEEK_C (AUTOSEEK_AT_MIN - (AUTOSEEK_K) * (AUTOSEQ_TEMPO_LOW)) + + #define CHECK_LIMITS(x, mi, ma) (((x) < (mi)) ? (mi) : (((x) > (ma)) ? (ma) : (x))) + + double seq, seek; + + if (bAutoSeqSetting) + { + seq = AUTOSEQ_C + AUTOSEQ_K * tempo; + seq = CHECK_LIMITS(seq, AUTOSEQ_AT_MAX, AUTOSEQ_AT_MIN); + sequenceMs = (int)(seq + 0.5); + } + + if (bAutoSeekSetting) + { + seek = AUTOSEEK_C + AUTOSEEK_K * tempo; + seek = CHECK_LIMITS(seek, AUTOSEEK_AT_MAX, AUTOSEEK_AT_MIN); + seekWindowMs = (int)(seek + 0.5); + } + + // Update seek window lengths + seekWindowLength = (sampleRate * sequenceMs) / 1000; + if (seekWindowLength < 2 * overlapLength) + { + seekWindowLength = 2 * overlapLength; + } + seekLength = (sampleRate * seekWindowMs) / 1000; +} + + + +// Sets new target tempo. Normal tempo = 'SCALE', smaller values represent slower +// tempo, larger faster tempo. +void TDStretch::setTempo(double newTempo) +{ + int intskip; + + tempo = newTempo; + + // Calculate new sequence duration + calcSeqParameters(); + + // Calculate ideal skip length (according to tempo value) + nominalSkip = tempo * (seekWindowLength - overlapLength); + intskip = (int)(nominalSkip + 0.5); + + // Calculate how many samples are needed in the 'inputBuffer' to + // process another batch of samples + //sampleReq = max(intskip + overlapLength, seekWindowLength) + seekLength / 2; + sampleReq = max(intskip + overlapLength, seekWindowLength) + seekLength; +} + + + +// Sets the number of channels, 1 = mono, 2 = stereo +void TDStretch::setChannels(int numChannels) +{ + assert(numChannels > 0); + if (channels == numChannels) return; +// assert(numChannels == 1 || numChannels == 2); + + channels = numChannels; + inputBuffer.setChannels(channels); + outputBuffer.setChannels(channels); + + // re-init overlap/buffer + overlapLength=0; + setParameters(sampleRate); +} + + +// nominal tempo, no need for processing, just pass the samples through +// to outputBuffer +/* +void TDStretch::processNominalTempo() +{ + assert(tempo == 1.0f); + + if (bMidBufferDirty) + { + // If there are samples in pMidBuffer waiting for overlapping, + // do a single sliding overlapping with them in order to prevent a + // clicking distortion in the output sound + if (inputBuffer.numSamples() < overlapLength) + { + // wait until we've got overlapLength input samples + return; + } + // Mix the samples in the beginning of 'inputBuffer' with the + // samples in 'midBuffer' using sliding overlapping + overlap(outputBuffer.ptrEnd(overlapLength), inputBuffer.ptrBegin(), 0); + outputBuffer.putSamples(overlapLength); + inputBuffer.receiveSamples(overlapLength); + clearMidBuffer(); + // now we've caught the nominal sample flow and may switch to + // bypass mode + } + + // Simply bypass samples from input to output + outputBuffer.moveSamples(inputBuffer); +} +*/ + + +// Processes as many processing frames of the samples 'inputBuffer', store +// the result into 'outputBuffer' +void TDStretch::processSamples() +{ + int ovlSkip, offset; + int temp; + + /* Removed this small optimization - can introduce a click to sound when tempo setting + crosses the nominal value + if (tempo == 1.0f) + { + // tempo not changed from the original, so bypass the processing + processNominalTempo(); + return; + } + */ + + // Process samples as long as there are enough samples in 'inputBuffer' + // to form a processing frame. + while ((int)inputBuffer.numSamples() >= sampleReq) + { + // If tempo differs from the normal ('SCALE'), scan for the best overlapping + // position + offset = seekBestOverlapPosition(inputBuffer.ptrBegin()); + + // Mix the samples in the 'inputBuffer' at position of 'offset' with the + // samples in 'midBuffer' using sliding overlapping + // ... first partially overlap with the end of the previous sequence + // (that's in 'midBuffer') + overlap(outputBuffer.ptrEnd((uint)overlapLength), inputBuffer.ptrBegin(), (uint)offset); + outputBuffer.putSamples((uint)overlapLength); + + // ... then copy sequence samples from 'inputBuffer' to output: + + // length of sequence + temp = (seekWindowLength - 2 * overlapLength); + + // crosscheck that we don't have buffer overflow... + if ((int)inputBuffer.numSamples() < (offset + temp + overlapLength * 2)) + { + continue; // just in case, shouldn't really happen + } + + outputBuffer.putSamples(inputBuffer.ptrBegin() + channels * (offset + overlapLength), (uint)temp); + + // Copies the end of the current sequence from 'inputBuffer' to + // 'midBuffer' for being mixed with the beginning of the next + // processing sequence and so on + assert((offset + temp + overlapLength * 2) <= (int)inputBuffer.numSamples()); + memcpy(pMidBuffer, inputBuffer.ptrBegin() + channels * (offset + temp + overlapLength), + channels * sizeof(SAMPLETYPE) * overlapLength); + + // Remove the processed samples from the input buffer. Update + // the difference between integer & nominal skip step to 'skipFract' + // in order to prevent the error from accumulating over time. + skipFract += nominalSkip; // real skip size + ovlSkip = (int)skipFract; // rounded to integer skip + skipFract -= ovlSkip; // maintain the fraction part, i.e. real vs. integer skip + inputBuffer.receiveSamples((uint)ovlSkip); + } +} + + +// Adds 'numsamples' pcs of samples from the 'samples' memory position into +// the input of the object. +void TDStretch::putSamples(const SAMPLETYPE *samples, uint nSamples) +{ + // Add the samples into the input buffer + inputBuffer.putSamples(samples, nSamples); + // Process the samples in input buffer + processSamples(); +} + + + +/// Set new overlap length parameter & reallocate RefMidBuffer if necessary. +void TDStretch::acceptNewOverlapLength(int newOverlapLength) +{ + int prevOvl; + + assert(newOverlapLength >= 0); + prevOvl = overlapLength; + overlapLength = newOverlapLength; + + if (overlapLength > prevOvl) + { + delete[] pMidBufferUnaligned; + + pMidBufferUnaligned = new SAMPLETYPE[overlapLength * channels + 16 / sizeof(SAMPLETYPE)]; + // ensure that 'pMidBuffer' is aligned to 16 byte boundary for efficiency + pMidBuffer = (SAMPLETYPE *)SOUNDTOUCH_ALIGN_POINTER_16(pMidBufferUnaligned); + + clearMidBuffer(); + } +} + + +// Operator 'new' is overloaded so that it automatically creates a suitable instance +// depending on if we've a MMX/SSE/etc-capable CPU available or not. +void * TDStretch::operator new(size_t s) +{ + // Notice! don't use "new TDStretch" directly, use "newInstance" to create a new instance instead! + ST_THROW_RT_ERROR("Error in TDStretch::new: Don't use 'new TDStretch' directly, use 'newInstance' member instead!"); + return newInstance(); +} + + +TDStretch * TDStretch::newInstance() +{ + uint uExtensions; + + uExtensions = detectCPUextensions(); + + // Check if MMX/SSE instruction set extensions supported by CPU + +#ifdef SOUNDTOUCH_ALLOW_MMX + // MMX routines available only with integer sample types + if (uExtensions & SUPPORT_MMX) + { + return ::new TDStretchMMX; + } + else +#endif // SOUNDTOUCH_ALLOW_MMX + + +#ifdef SOUNDTOUCH_ALLOW_SSE + if (uExtensions & SUPPORT_SSE) + { + // SSE support + return ::new TDStretchSSE; + } + else +#endif // SOUNDTOUCH_ALLOW_SSE + + { + // ISA optimizations not supported, use plain C version + return ::new TDStretch; + } +} + + +////////////////////////////////////////////////////////////////////////////// +// +// Integer arithmetics specific algorithm implementations. +// +////////////////////////////////////////////////////////////////////////////// + +#ifdef SOUNDTOUCH_INTEGER_SAMPLES + +// Overlaps samples in 'midBuffer' with the samples in 'input'. The 'Stereo' +// version of the routine. +void TDStretch::overlapStereo(short *poutput, const short *input) const +{ + int i; + short temp; + int cnt2; + + for (i = 0; i < overlapLength ; i ++) + { + temp = (short)(overlapLength - i); + cnt2 = 2 * i; + poutput[cnt2] = (input[cnt2] * i + pMidBuffer[cnt2] * temp ) / overlapLength; + poutput[cnt2 + 1] = (input[cnt2 + 1] * i + pMidBuffer[cnt2 + 1] * temp ) / overlapLength; + } +} + + +// Overlaps samples in 'midBuffer' with the samples in 'input'. The 'Multi' +// version of the routine. +void TDStretch::overlapMulti(SAMPLETYPE *poutput, const SAMPLETYPE *input) const +{ + SAMPLETYPE m1=(SAMPLETYPE)0; + SAMPLETYPE m2; + int i=0; + + for (m2 = (SAMPLETYPE)overlapLength; m2; m2 --) + { + for (int c = 0; c < channels; c ++) + { + poutput[i] = (input[i] * m1 + pMidBuffer[i] * m2) / overlapLength; + i++; + } + + m1++; + } +} + +// Calculates the x having the closest 2^x value for the given value +static int _getClosest2Power(double value) +{ + return (int)(log(value) / log(2.0) + 0.5); +} + + +/// Calculates overlap period length in samples. +/// Integer version rounds overlap length to closest power of 2 +/// for a divide scaling operation. +void TDStretch::calculateOverlapLength(int aoverlapMs) +{ + int newOvl; + + assert(aoverlapMs >= 0); + + // calculate overlap length so that it's power of 2 - thus it's easy to do + // integer division by right-shifting. Term "-1" at end is to account for + // the extra most significatnt bit left unused in result by signed multiplication + overlapDividerBitsPure = _getClosest2Power((sampleRate * aoverlapMs) / 1000.0) - 1; + if (overlapDividerBitsPure > 9) overlapDividerBitsPure = 9; + if (overlapDividerBitsPure < 3) overlapDividerBitsPure = 3; + newOvl = (int)pow(2.0, (int)overlapDividerBitsPure + 1); // +1 => account for -1 above + + acceptNewOverlapLength(newOvl); + + overlapDividerBitsNorm = overlapDividerBitsPure; + + // calculate sloping divider so that crosscorrelation operation won't + // overflow 32-bit register. Max. sum of the crosscorrelation sum without + // divider would be 2^30*(N^3-N)/3, where N = overlap length + slopingDivider = (newOvl * newOvl - 1) / 3; +} + + +double TDStretch::calcCrossCorr(const short *mixingPos, const short *compare, double &norm) +{ + long corr; + unsigned long lnorm; + int i; + + corr = lnorm = 0; + // Same routine for stereo and mono. For stereo, unroll loop for better + // efficiency and gives slightly better resolution against rounding. + // For mono it same routine, just unrolls loop by factor of 4 + for (i = 0; i < channels * overlapLength; i += 4) + { + corr += (mixingPos[i] * compare[i] + + mixingPos[i + 1] * compare[i + 1]) >> overlapDividerBitsNorm; // notice: do intermediate division here to avoid integer overflow + corr += (mixingPos[i + 2] * compare[i + 2] + + mixingPos[i + 3] * compare[i + 3]) >> overlapDividerBitsNorm; + lnorm += (mixingPos[i] * mixingPos[i] + + mixingPos[i + 1] * mixingPos[i + 1]) >> overlapDividerBitsNorm; // notice: do intermediate division here to avoid integer overflow + lnorm += (mixingPos[i + 2] * mixingPos[i + 2] + + mixingPos[i + 3] * mixingPos[i + 3]) >> overlapDividerBitsNorm; + } + + if (lnorm > maxnorm) + { + maxnorm = lnorm; + } + // Normalize result by dividing by sqrt(norm) - this step is easiest + // done using floating point operation + norm = (double)lnorm; + return (double)corr / sqrt((norm < 1e-9) ? 1.0 : norm); +} + + +/// Update cross-correlation by accumulating "norm" coefficient by previously calculated value +double TDStretch::calcCrossCorrAccumulate(const short *mixingPos, const short *compare, double &norm) +{ + long corr; + unsigned long lnorm; + int i; + + // cancel first normalizer tap from previous round + lnorm = 0; + for (i = 1; i <= channels; i ++) + { + lnorm -= (mixingPos[-i] * mixingPos[-i]) >> overlapDividerBitsNorm; + } + + corr = 0; + // Same routine for stereo and mono. For stereo, unroll loop for better + // efficiency and gives slightly better resolution against rounding. + // For mono it same routine, just unrolls loop by factor of 4 + for (i = 0; i < channels * overlapLength; i += 4) + { + corr += (mixingPos[i] * compare[i] + + mixingPos[i + 1] * compare[i + 1]) >> overlapDividerBitsNorm; // notice: do intermediate division here to avoid integer overflow + corr += (mixingPos[i + 2] * compare[i + 2] + + mixingPos[i + 3] * compare[i + 3]) >> overlapDividerBitsNorm; + } + + // update normalizer with last samples of this round + for (int j = 0; j < channels; j ++) + { + i --; + lnorm += (mixingPos[i] * mixingPos[i]) >> overlapDividerBitsNorm; + } + + norm += (double)lnorm; + if (norm > maxnorm) + { + maxnorm = (unsigned long)norm; + } + + // Normalize result by dividing by sqrt(norm) - this step is easiest + // done using floating point operation + return (double)corr / sqrt((norm < 1e-9) ? 1.0 : norm); +} + +#endif // SOUNDTOUCH_INTEGER_SAMPLES + +////////////////////////////////////////////////////////////////////////////// +// +// Floating point arithmetics specific algorithm implementations. +// + +#ifdef SOUNDTOUCH_FLOAT_SAMPLES + +// Overlaps samples in 'midBuffer' with the samples in 'pInput' +void TDStretch::overlapStereo(float *pOutput, const float *pInput) const +{ + int i; + float fScale; + float f1; + float f2; + + fScale = 1.0f / (float)overlapLength; + + f1 = 0; + f2 = 1.0f; + + for (i = 0; i < 2 * (int)overlapLength ; i += 2) + { + pOutput[i + 0] = pInput[i + 0] * f1 + pMidBuffer[i + 0] * f2; + pOutput[i + 1] = pInput[i + 1] * f1 + pMidBuffer[i + 1] * f2; + + f1 += fScale; + f2 -= fScale; + } +} + + +// Overlaps samples in 'midBuffer' with the samples in 'input'. +void TDStretch::overlapMulti(float *pOutput, const float *pInput) const +{ + int i; + float fScale; + float f1; + float f2; + + fScale = 1.0f / (float)overlapLength; + + f1 = 0; + f2 = 1.0f; + + i=0; + for (int i2 = 0; i2 < overlapLength; i2 ++) + { + // note: Could optimize this slightly by taking into account that always channels > 2 + for (int c = 0; c < channels; c ++) + { + pOutput[i] = pInput[i] * f1 + pMidBuffer[i] * f2; + i++; + } + f1 += fScale; + f2 -= fScale; + } +} + + +/// Calculates overlapInMsec period length in samples. +void TDStretch::calculateOverlapLength(int overlapInMsec) +{ + int newOvl; + + assert(overlapInMsec >= 0); + newOvl = (sampleRate * overlapInMsec) / 1000; + if (newOvl < 16) newOvl = 16; + + // must be divisible by 8 + newOvl -= newOvl % 8; + + acceptNewOverlapLength(newOvl); +} + + +/// Calculate cross-correlation +double TDStretch::calcCrossCorr(const float *mixingPos, const float *compare, double &anorm) +{ + double corr; + double norm; + int i; + + corr = norm = 0; + // Same routine for stereo and mono. For Stereo, unroll by factor of 2. + // For mono it's same routine yet unrollsd by factor of 4. + for (i = 0; i < channels * overlapLength; i += 4) + { + corr += mixingPos[i] * compare[i] + + mixingPos[i + 1] * compare[i + 1]; + + norm += mixingPos[i] * mixingPos[i] + + mixingPos[i + 1] * mixingPos[i + 1]; + + // unroll the loop for better CPU efficiency: + corr += mixingPos[i + 2] * compare[i + 2] + + mixingPos[i + 3] * compare[i + 3]; + + norm += mixingPos[i + 2] * mixingPos[i + 2] + + mixingPos[i + 3] * mixingPos[i + 3]; + } + + anorm = norm; + return corr / sqrt((norm < 1e-9 ? 1.0 : norm)); +} + + +/// Update cross-correlation by accumulating "norm" coefficient by previously calculated value +double TDStretch::calcCrossCorrAccumulate(const float *mixingPos, const float *compare, double &norm) +{ + double corr; + int i; + + corr = 0; + + // cancel first normalizer tap from previous round + for (i = 1; i <= channels; i ++) + { + norm -= mixingPos[-i] * mixingPos[-i]; + } + + // Same routine for stereo and mono. For Stereo, unroll by factor of 2. + // For mono it's same routine yet unrollsd by factor of 4. + for (i = 0; i < channels * overlapLength; i += 4) + { + corr += mixingPos[i] * compare[i] + + mixingPos[i + 1] * compare[i + 1] + + mixingPos[i + 2] * compare[i + 2] + + mixingPos[i + 3] * compare[i + 3]; + } + + // update normalizer with last samples of this round + for (int j = 0; j < channels; j ++) + { + i --; + norm += mixingPos[i] * mixingPos[i]; + } + + return corr / sqrt((norm < 1e-9 ? 1.0 : norm)); +} + + +#endif // SOUNDTOUCH_FLOAT_SAMPLES diff --git a/src/TDStretch.h b/src/TDStretch.h new file mode 100644 index 0000000..e6d75aa --- /dev/null +++ b/src/TDStretch.h @@ -0,0 +1,281 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// Sampled sound tempo changer/time stretch algorithm. Changes the sound tempo +/// while maintaining the original pitch by using a time domain WSOLA-like method +/// with several performance-increasing tweaks. +/// +/// Note : MMX/SSE optimized functions reside in separate, platform-specific files +/// 'mmx_optimized.cpp' and 'sse_optimized.cpp' +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2015-08-09 00:00:15 +0300 (Sun, 09 Aug 2015) $ +// File revision : $Revision: 4 $ +// +// $Id: TDStretch.h 226 2015-08-08 21:00:15Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#ifndef TDStretch_H +#define TDStretch_H + +#include +#include "STTypes.h" +#include "RateTransposer.h" +#include "FIFOSamplePipe.h" + +namespace soundtouch +{ + +/// Default values for sound processing parameters: +/// Notice that the default parameters are tuned for contemporary popular music +/// processing. For speech processing applications these parameters suit better: +/// #define DEFAULT_SEQUENCE_MS 40 +/// #define DEFAULT_SEEKWINDOW_MS 15 +/// #define DEFAULT_OVERLAP_MS 8 +/// + +/// Default length of a single processing sequence, in milliseconds. This determines to how +/// long sequences the original sound is chopped in the time-stretch algorithm. +/// +/// The larger this value is, the lesser sequences are used in processing. In principle +/// a bigger value sounds better when slowing down tempo, but worse when increasing tempo +/// and vice versa. +/// +/// Increasing this value reduces computational burden & vice versa. +//#define DEFAULT_SEQUENCE_MS 40 +#define DEFAULT_SEQUENCE_MS USE_AUTO_SEQUENCE_LEN + +/// Giving this value for the sequence length sets automatic parameter value +/// according to tempo setting (recommended) +#define USE_AUTO_SEQUENCE_LEN 0 + +/// Seeking window default length in milliseconds for algorithm that finds the best possible +/// overlapping location. This determines from how wide window the algorithm may look for an +/// optimal joining location when mixing the sound sequences back together. +/// +/// The bigger this window setting is, the higher the possibility to find a better mixing +/// position will become, but at the same time large values may cause a "drifting" artifact +/// because consequent sequences will be taken at more uneven intervals. +/// +/// If there's a disturbing artifact that sounds as if a constant frequency was drifting +/// around, try reducing this setting. +/// +/// Increasing this value increases computational burden & vice versa. +//#define DEFAULT_SEEKWINDOW_MS 15 +#define DEFAULT_SEEKWINDOW_MS USE_AUTO_SEEKWINDOW_LEN + +/// Giving this value for the seek window length sets automatic parameter value +/// according to tempo setting (recommended) +#define USE_AUTO_SEEKWINDOW_LEN 0 + +/// Overlap length in milliseconds. When the chopped sound sequences are mixed back together, +/// to form a continuous sound stream, this parameter defines over how long period the two +/// consecutive sequences are let to overlap each other. +/// +/// This shouldn't be that critical parameter. If you reduce the DEFAULT_SEQUENCE_MS setting +/// by a large amount, you might wish to try a smaller value on this. +/// +/// Increasing this value increases computational burden & vice versa. +#define DEFAULT_OVERLAP_MS 8 + + +/// Class that does the time-stretch (tempo change) effect for the processed +/// sound. +class TDStretch : public FIFOProcessor +{ +protected: + int channels; + int sampleReq; + + int overlapLength; + int seekLength; + int seekWindowLength; + int overlapDividerBitsNorm; + int overlapDividerBitsPure; + int slopingDivider; + int sampleRate; + int sequenceMs; + int seekWindowMs; + int overlapMs; + + unsigned long maxnorm; + float maxnormf; + + double tempo; + double nominalSkip; + double skipFract; + + bool bQuickSeek; + bool bAutoSeqSetting; + bool bAutoSeekSetting; + + SAMPLETYPE *pMidBuffer; + SAMPLETYPE *pMidBufferUnaligned; + + FIFOSampleBuffer outputBuffer; + FIFOSampleBuffer inputBuffer; + + void acceptNewOverlapLength(int newOverlapLength); + + virtual void clearCrossCorrState(); + void calculateOverlapLength(int overlapMs); + + virtual double calcCrossCorr(const SAMPLETYPE *mixingPos, const SAMPLETYPE *compare, double &norm); + virtual double calcCrossCorrAccumulate(const SAMPLETYPE *mixingPos, const SAMPLETYPE *compare, double &norm); + + virtual int seekBestOverlapPositionFull(const SAMPLETYPE *refPos); + virtual int seekBestOverlapPositionQuick(const SAMPLETYPE *refPos); + virtual int seekBestOverlapPosition(const SAMPLETYPE *refPos); + + virtual void overlapStereo(SAMPLETYPE *output, const SAMPLETYPE *input) const; + virtual void overlapMono(SAMPLETYPE *output, const SAMPLETYPE *input) const; + virtual void overlapMulti(SAMPLETYPE *output, const SAMPLETYPE *input) const; + + void clearMidBuffer(); + void overlap(SAMPLETYPE *output, const SAMPLETYPE *input, uint ovlPos) const; + + void calcSeqParameters(); + void adaptNormalizer(); + + + /// Changes the tempo of the given sound samples. + /// Returns amount of samples returned in the "output" buffer. + /// The maximum amount of samples that can be returned at a time is set by + /// the 'set_returnBuffer_size' function. + void processSamples(); + +public: + TDStretch(); + virtual ~TDStretch(); + + /// Operator 'new' is overloaded so that it automatically creates a suitable instance + /// depending on if we've a MMX/SSE/etc-capable CPU available or not. + static void *operator new(size_t s); + + /// Use this function instead of "new" operator to create a new instance of this class. + /// This function automatically chooses a correct feature set depending on if the CPU + /// supports MMX/SSE/etc extensions. + static TDStretch *newInstance(); + + /// Returns the output buffer object + FIFOSamplePipe *getOutput() { return &outputBuffer; }; + + /// Returns the input buffer object + FIFOSamplePipe *getInput() { return &inputBuffer; }; + + /// Sets new target tempo. Normal tempo = 'SCALE', smaller values represent slower + /// tempo, larger faster tempo. + void setTempo(double newTempo); + + /// Returns nonzero if there aren't any samples available for outputting. + virtual void clear(); + + /// Clears the input buffer + void clearInput(); + + /// Sets the number of channels, 1 = mono, 2 = stereo + void setChannels(int numChannels); + + /// Enables/disables the quick position seeking algorithm. Zero to disable, + /// nonzero to enable + void enableQuickSeek(bool enable); + + /// Returns nonzero if the quick seeking algorithm is enabled. + bool isQuickSeekEnabled() const; + + /// Sets routine control parameters. These control are certain time constants + /// defining how the sound is stretched to the desired duration. + // + /// 'sampleRate' = sample rate of the sound + /// 'sequenceMS' = one processing sequence length in milliseconds + /// 'seekwindowMS' = seeking window length for scanning the best overlapping + /// position + /// 'overlapMS' = overlapping length + void setParameters(int sampleRate, ///< Samplerate of sound being processed (Hz) + int sequenceMS = -1, ///< Single processing sequence length (ms) + int seekwindowMS = -1, ///< Offset seeking window length (ms) + int overlapMS = -1 ///< Sequence overlapping length (ms) + ); + + /// Get routine control parameters, see setParameters() function. + /// Any of the parameters to this function can be NULL, in such case corresponding parameter + /// value isn't returned. + void getParameters(int *pSampleRate, int *pSequenceMs, int *pSeekWindowMs, int *pOverlapMs) const; + + /// Adds 'numsamples' pcs of samples from the 'samples' memory position into + /// the input of the object. + virtual void putSamples( + const SAMPLETYPE *samples, ///< Input sample data + uint numSamples ///< Number of samples in 'samples' so that one sample + ///< contains both channels if stereo + ); + + /// return nominal input sample requirement for triggering a processing batch + int getInputSampleReq() const + { + return (int)(nominalSkip + 0.5); + } + + /// return nominal output sample amount when running a processing batch + int getOutputBatchSize() const + { + return seekWindowLength - overlapLength; + } +}; + + + +// Implementation-specific class declarations: + +#ifdef SOUNDTOUCH_ALLOW_MMX + /// Class that implements MMX optimized routines for 16bit integer samples type. + class TDStretchMMX : public TDStretch + { + protected: + double calcCrossCorr(const short *mixingPos, const short *compare, double &norm); + double calcCrossCorrAccumulate(const short *mixingPos, const short *compare, double &norm); + virtual void overlapStereo(short *output, const short *input) const; + virtual void clearCrossCorrState(); + }; +#endif /// SOUNDTOUCH_ALLOW_MMX + + +#ifdef SOUNDTOUCH_ALLOW_SSE + /// Class that implements SSE optimized routines for floating point samples type. + class TDStretchSSE : public TDStretch + { + protected: + double calcCrossCorr(const float *mixingPos, const float *compare, double &norm); + double calcCrossCorrAccumulate(const float *mixingPos, const float *compare, double &norm); + }; + +#endif /// SOUNDTOUCH_ALLOW_SSE + +} +#endif /// TDStretch_H diff --git a/src/cpu_detect.h b/src/cpu_detect.h new file mode 100644 index 0000000..7859ffb --- /dev/null +++ b/src/cpu_detect.h @@ -0,0 +1,62 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// A header file for detecting the Intel MMX instructions set extension. +/// +/// Please see 'mmx_win.cpp', 'mmx_cpp.cpp' and 'mmx_non_x86.cpp' for the +/// routine implementations for x86 Windows, x86 gnu version and non-x86 +/// platforms, respectively. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2008-02-10 18:26:55 +0200 (Sun, 10 Feb 2008) $ +// File revision : $Revision: 4 $ +// +// $Id: cpu_detect.h 11 2008-02-10 16:26:55Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#ifndef _CPU_DETECT_H_ +#define _CPU_DETECT_H_ + +#include "STTypes.h" + +#define SUPPORT_MMX 0x0001 +#define SUPPORT_3DNOW 0x0002 +#define SUPPORT_ALTIVEC 0x0004 +#define SUPPORT_SSE 0x0008 +#define SUPPORT_SSE2 0x0010 + +/// Checks which instruction set extensions are supported by the CPU. +/// +/// \return A bitmask of supported extensions, see SUPPORT_... defines. +uint detectCPUextensions(void); + +/// Disables given set of instruction extensions. See SUPPORT_... defines. +void disableExtensions(uint wDisableMask); + +#endif // _CPU_DETECT_H_ diff --git a/src/cpu_detect_x86.cpp b/src/cpu_detect_x86.cpp new file mode 100644 index 0000000..00f22ab --- /dev/null +++ b/src/cpu_detect_x86.cpp @@ -0,0 +1,138 @@ +//////////////////////////////////////////////////////////////////////////////// +/// +/// Generic version of the x86 CPU extension detection routine. +/// +/// This file is for GNU & other non-Windows compilers, see 'cpu_detect_x86_win.cpp' +/// for the Microsoft compiler version. +/// +/// Author : Copyright (c) Olli Parviainen +/// Author e-mail : oparviai 'at' iki.fi +/// SoundTouch WWW: http://www.surina.net/soundtouch +/// +//////////////////////////////////////////////////////////////////////////////// +// +// Last changed : $Date: 2014-01-07 20:24:28 +0200 (Tue, 07 Jan 2014) $ +// File revision : $Revision: 4 $ +// +// $Id: cpu_detect_x86.cpp 183 2014-01-07 18:24:28Z oparviai $ +// +//////////////////////////////////////////////////////////////////////////////// +// +// License : +// +// SoundTouch audio processing library +// Copyright (c) Olli Parviainen +// +// This library is free software; you can redistribute it and/or +// modify it under the terms of the GNU Lesser General Public +// License as published by the Free Software Foundation; either +// version 2.1 of the License, or (at your option) any later version. +// +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +// Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public +// License along with this library; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +// +//////////////////////////////////////////////////////////////////////////////// + +#include "cpu_detect.h" +#include "STTypes.h" + + +#if defined(SOUNDTOUCH_ALLOW_X86_OPTIMIZATIONS) + + #if defined(__GNUC__) && defined(__i386__) + // gcc + #include "cpuid.h" + #elif defined(_M_IX86) + // windows non-gcc + #include + #endif + + #define bit_MMX (1 << 23) + #define bit_SSE (1 << 25) + #define bit_SSE2 (1 << 26) +#endif + + +////////////////////////////////////////////////////////////////////////////// +// +// processor instructions extension detection routines +// +////////////////////////////////////////////////////////////////////////////// + +// Flag variable indicating whick ISA extensions are disabled (for debugging) +static uint _dwDisabledISA = 0x00; // 0xffffffff; //<- use this to disable all extensions + +// Disables given set of instruction extensions. See SUPPORT_... defines. +void disableExtensions(uint dwDisableMask) +{ + _dwDisabledISA = dwDisableMask; +} + + + +/// Checks which instruction set extensions are supported by the CPU. +uint detectCPUextensions(void) +{ +/// If building for a 64bit system (no Itanium) and the user wants optimizations. +/// Return the OR of SUPPORT_{MMX,SSE,SSE2}. 11001 or 0x19. +/// Keep the _dwDisabledISA test (2 more operations, could be eliminated). +#if ((defined(__GNUC__) && defined(__x86_64__)) \ + || defined(_M_X64)) \ + && defined(SOUNDTOUCH_ALLOW_X86_OPTIMIZATIONS) + return 0x19 & ~_dwDisabledISA; + +/// If building for a 32bit system and the user wants optimizations. +/// Keep the _dwDisabledISA test (2 more operations, could be eliminated). +#elif ((defined(__GNUC__) && defined(__i386__)) \ + || defined(_M_IX86)) \ + && defined(SOUNDTOUCH_ALLOW_X86_OPTIMIZATIONS) + + if (_dwDisabledISA == 0xffffffff) return 0; + + uint res = 0; + +#if defined(__GNUC__) + // GCC version of cpuid. Requires GCC 4.3.0 or later for __cpuid intrinsic support. + uint eax, ebx, ecx, edx; // unsigned int is the standard type. uint is defined by the compiler and not guaranteed to be portable. + + // Check if no cpuid support. + if (!__get_cpuid (1, &eax, &ebx, &ecx, &edx)) return 0; // always disable extensions. + + if (edx & bit_MMX) res = res | SUPPORT_MMX; + if (edx & bit_SSE) res = res | SUPPORT_SSE; + if (edx & bit_SSE2) res = res | SUPPORT_SSE2; + +#else + // Window / VS version of cpuid. Notice that Visual Studio 2005 or later required + // for __cpuid intrinsic support. + int reg[4] = {-1}; + + // Check if no cpuid support. + __cpuid(reg,0); + if ((unsigned int)reg[0] == 0) return 0; // always disable extensions. + + __cpuid(reg,1); + if ((unsigned int)reg[3] & bit_MMX) res = res | SUPPORT_MMX; + if ((unsigned int)reg[3] & bit_SSE) res = res | SUPPORT_SSE; + if ((unsigned int)reg[3] & bit_SSE2) res = res | SUPPORT_SSE2; + +#endif + + return res & ~_dwDisabledISA; + +#else + +/// One of these is true: +/// 1) We don't want optimizations. +/// 2) Using an unsupported compiler. +/// 3) Running on a non-x86 platform. + return 0; + +#endif +}