Chris@0: /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */ Chris@0: Chris@0: /* Chris@0: Sonic Visualiser Chris@0: An audio file viewer and annotation editor. Chris@0: Centre for Digital Music, Queen Mary, University of London. Chris@0: This file copyright 2006 Chris Cannam. Chris@0: Chris@0: This program is free software; you can redistribute it and/or Chris@0: modify it under the terms of the GNU General Public License as Chris@0: published by the Free Software Foundation; either version 2 of the Chris@0: License, or (at your option) any later version. See the file Chris@0: COPYING included with this distribution for more information. Chris@0: */ Chris@0: Chris@14: #ifndef _PHASE_VOCODER_TIME_STRETCHER_H_ Chris@14: #define _PHASE_VOCODER_TIME_STRETCHER_H_ Chris@0: Chris@0: #include "base/Window.h" Chris@0: #include "base/RingBuffer.h" Chris@0: Chris@0: #include Chris@0: Chris@0: /** Chris@14: * A time stretcher that alters the performance speed of audio, Chris@22: * preserving pitch. Chris@0: * Chris@22: * This is based on the straightforward phase vocoder with phase Chris@22: * unwrapping (as in e.g. the DAFX book pp275-), with optional Chris@22: * percussive transient detection to avoid smearing percussive notes Chris@22: * and resynchronise phases, and adding a stream API for real-time Chris@22: * use. Principles and methods from Chris Duxbury, AES 2002 and 2004 Chris@22: * thesis; Emmanuel Ravelli, DAFX 2005; Dan Barry, ISSC 2005 on Chris@22: * percussion detection; code by Chris Cannam. Chris@0: */ Chris@0: Chris@14: class PhaseVocoderTimeStretcher Chris@0: { Chris@0: public: Chris@22: PhaseVocoderTimeStretcher(size_t sampleRate, Chris@22: size_t channels, Chris@16: float ratio, Chris@16: bool sharpen, Chris@16: size_t maxProcessInputBlockSize); Chris@14: virtual ~PhaseVocoderTimeStretcher(); Chris@0: Chris@12: /** Chris@12: * Process a block. The input array contains the given number of Chris@16: * samples (on each channel); the output must have space for Chris@16: * lrintf(samples * m_ratio). Chris@16: * Chris@16: * This should work correctly for some ratios, e.g. small powers Chris@16: * of two. For other ratios it may drop samples -- use putInput Chris@16: * in a loop followed by getOutput (when getAvailableOutputSamples Chris@16: * reports enough) instead. Chris@16: * Chris@16: * Do not mix process calls with putInput/getOutput calls. Chris@12: */ Chris@16: void process(float **input, float **output, size_t samples); Chris@16: Chris@16: /** Chris@16: * Return the number of samples that would need to be added via Chris@16: * putInput in order to provoke the time stretcher into doing some Chris@16: * time stretching and making more output samples available. Chris@16: */ Chris@16: size_t getRequiredInputSamples() const; Chris@16: Chris@16: /** Chris@16: * Put (and possibly process) a given number of input samples. Chris@16: * Number must not exceed the maxProcessInputBlockSize passed to Chris@16: * constructor. Chris@16: */ Chris@16: void putInput(float **input, size_t samples); Chris@16: Chris@16: size_t getAvailableOutputSamples() const; Chris@16: Chris@16: void getOutput(float **output, size_t samples); Chris@16: Chris@16: //!!! and reset? Chris@0: Chris@0: /** Chris@15: * Get the hop size for input. Chris@0: */ Chris@0: size_t getInputIncrement() const { return m_n1; } Chris@0: Chris@0: /** Chris@15: * Get the hop size for output. Chris@15: */ Chris@16: size_t getOutputIncrement() const { return m_n2; } Chris@15: Chris@15: /** Chris@15: * Get the window size for FFT processing. Chris@0: */ Chris@0: size_t getWindowSize() const { return m_wlen; } Chris@0: Chris@0: /** Chris@15: * Get the window type. Chris@0: */ Chris@20: // WindowType getWindowType() const { return m_window->getType(); } Chris@0: Chris@15: /** Chris@16: * Get the stretch ratio. Chris@15: */ Chris@16: float getRatio() const { return float(m_n2) / float(m_n1); } Chris@16: Chris@16: /** Chris@16: * Return whether this time stretcher will attempt to sharpen transients. Chris@16: */ Chris@16: bool getSharpening() const { return m_sharpen; } Chris@15: Chris@15: /** Chris@15: * Get the latency added by the time stretcher, in sample frames. Chris@15: */ Chris@0: size_t getProcessingLatency() const; Chris@0: Chris@0: protected: Chris@13: /** Chris@13: * Process a single phase vocoder frame. Chris@13: * Chris@13: * Take m_wlen time-domain source samples from in, perform an FFT, Chris@13: * phase shift, and IFFT, and add the results to out (presumably Chris@13: * overlapping parts of existing data from prior frames). Chris@13: * Chris@13: * Also add to the modulation output the results of windowing a Chris@13: * set of 1s with the resynthesis window -- this can then be used Chris@13: * to ensure the output has the correct magnitude in cases where Chris@13: * the window overlap varies or otherwise results in something Chris@13: * other than a flat sum. Chris@13: */ Chris@20: Chris@20: Chris@20: void analyseBlock(size_t channel, float *in); // into m_freq[channel] Chris@20: Chris@20: bool isTransient(); // operates on m_freq[0..m_channels-1] Chris@20: Chris@20: void synthesiseBlock(size_t channel, float *out, float *modulation, Chris@20: size_t lastStep); Chris@0: Chris@22: size_t m_sampleRate; Chris@16: size_t m_channels; Chris@12: float m_ratio; Chris@16: bool m_sharpen; Chris@0: size_t m_n1; Chris@0: size_t m_n2; Chris@0: size_t m_wlen; Chris@20: Window *m_analysisWindow; Chris@20: Window *m_synthesisWindow; Chris@0: Chris@21: int m_totalCount; Chris@21: int m_transientCount; Chris@21: int m_n2sum; Chris@21: Chris@16: float **m_prevPhase; Chris@16: float **m_prevAdjustedPhase; Chris@16: Chris@20: float *m_prevTransientMag; Chris@21: int m_prevTransientScore; Chris@22: int m_transientThreshold; Chris@20: bool m_prevTransient; Chris@20: Chris@20: float *m_tempbuf; Chris@20: float **m_time; Chris@20: fftwf_complex **m_freq; Chris@20: fftwf_plan *m_plan; Chris@20: fftwf_plan *m_iplan; Chris@0: Chris@16: RingBuffer **m_inbuf; Chris@16: RingBuffer **m_outbuf; Chris@16: float **m_mashbuf; Chris@13: float *m_modulationbuf; Chris@0: }; Chris@0: Chris@0: #endif