1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libsoundtouch/src/TDStretch.h Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,272 @@
1.4 +////////////////////////////////////////////////////////////////////////////////
1.5 +///
1.6 +/// Sampled sound tempo changer/time stretch algorithm. Changes the sound tempo
1.7 +/// while maintaining the original pitch by using a time domain WSOLA-like method
1.8 +/// with several performance-increasing tweaks.
1.9 +///
1.10 +/// Note : MMX/SSE optimized functions reside in separate, platform-specific files
1.11 +/// 'mmx_optimized.cpp' and 'sse_optimized.cpp'
1.12 +///
1.13 +/// Author : Copyright (c) Olli Parviainen
1.14 +/// Author e-mail : oparviai 'at' iki.fi
1.15 +/// SoundTouch WWW: http://www.surina.net/soundtouch
1.16 +///
1.17 +////////////////////////////////////////////////////////////////////////////////
1.18 +//
1.19 +// Last changed : $Date: 2014-04-06 10:57:21 -0500 (Sun, 06 Apr 2014) $
1.20 +// File revision : $Revision: 4 $
1.21 +//
1.22 +// $Id: TDStretch.h 195 2014-04-06 15:57:21Z oparviai $
1.23 +//
1.24 +////////////////////////////////////////////////////////////////////////////////
1.25 +//
1.26 +// License :
1.27 +//
1.28 +// SoundTouch audio processing library
1.29 +// Copyright (c) Olli Parviainen
1.30 +//
1.31 +// This library is free software; you can redistribute it and/or
1.32 +// modify it under the terms of the GNU Lesser General Public
1.33 +// License as published by the Free Software Foundation; either
1.34 +// version 2.1 of the License, or (at your option) any later version.
1.35 +//
1.36 +// This library is distributed in the hope that it will be useful,
1.37 +// but WITHOUT ANY WARRANTY; without even the implied warranty of
1.38 +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
1.39 +// Lesser General Public License for more details.
1.40 +//
1.41 +// You should have received a copy of the GNU Lesser General Public
1.42 +// License along with this library; if not, write to the Free Software
1.43 +// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
1.44 +//
1.45 +////////////////////////////////////////////////////////////////////////////////
1.46 +
1.47 +#ifndef TDStretch_H
1.48 +#define TDStretch_H
1.49 +
1.50 +#include <stddef.h>
1.51 +#include "STTypes.h"
1.52 +#include "RateTransposer.h"
1.53 +#include "FIFOSamplePipe.h"
1.54 +
1.55 +namespace soundtouch
1.56 +{
1.57 +
1.58 +/// Default values for sound processing parameters:
1.59 +/// Notice that the default parameters are tuned for contemporary popular music
1.60 +/// processing. For speech processing applications these parameters suit better:
1.61 +/// #define DEFAULT_SEQUENCE_MS 40
1.62 +/// #define DEFAULT_SEEKWINDOW_MS 15
1.63 +/// #define DEFAULT_OVERLAP_MS 8
1.64 +///
1.65 +
1.66 +/// Default length of a single processing sequence, in milliseconds. This determines to how
1.67 +/// long sequences the original sound is chopped in the time-stretch algorithm.
1.68 +///
1.69 +/// The larger this value is, the lesser sequences are used in processing. In principle
1.70 +/// a bigger value sounds better when slowing down tempo, but worse when increasing tempo
1.71 +/// and vice versa.
1.72 +///
1.73 +/// Increasing this value reduces computational burden & vice versa.
1.74 +//#define DEFAULT_SEQUENCE_MS 40
1.75 +#define DEFAULT_SEQUENCE_MS USE_AUTO_SEQUENCE_LEN
1.76 +
1.77 +/// Giving this value for the sequence length sets automatic parameter value
1.78 +/// according to tempo setting (recommended)
1.79 +#define USE_AUTO_SEQUENCE_LEN 0
1.80 +
1.81 +/// Seeking window default length in milliseconds for algorithm that finds the best possible
1.82 +/// overlapping location. This determines from how wide window the algorithm may look for an
1.83 +/// optimal joining location when mixing the sound sequences back together.
1.84 +///
1.85 +/// The bigger this window setting is, the higher the possibility to find a better mixing
1.86 +/// position will become, but at the same time large values may cause a "drifting" artifact
1.87 +/// because consequent sequences will be taken at more uneven intervals.
1.88 +///
1.89 +/// If there's a disturbing artifact that sounds as if a constant frequency was drifting
1.90 +/// around, try reducing this setting.
1.91 +///
1.92 +/// Increasing this value increases computational burden & vice versa.
1.93 +//#define DEFAULT_SEEKWINDOW_MS 15
1.94 +#define DEFAULT_SEEKWINDOW_MS USE_AUTO_SEEKWINDOW_LEN
1.95 +
1.96 +/// Giving this value for the seek window length sets automatic parameter value
1.97 +/// according to tempo setting (recommended)
1.98 +#define USE_AUTO_SEEKWINDOW_LEN 0
1.99 +
1.100 +/// Overlap length in milliseconds. When the chopped sound sequences are mixed back together,
1.101 +/// to form a continuous sound stream, this parameter defines over how long period the two
1.102 +/// consecutive sequences are let to overlap each other.
1.103 +///
1.104 +/// This shouldn't be that critical parameter. If you reduce the DEFAULT_SEQUENCE_MS setting
1.105 +/// by a large amount, you might wish to try a smaller value on this.
1.106 +///
1.107 +/// Increasing this value increases computational burden & vice versa.
1.108 +#define DEFAULT_OVERLAP_MS 8
1.109 +
1.110 +
1.111 +/// Class that does the time-stretch (tempo change) effect for the processed
1.112 +/// sound.
1.113 +class TDStretch : public FIFOProcessor
1.114 +{
1.115 +protected:
1.116 + int channels;
1.117 + int sampleReq;
1.118 + float tempo;
1.119 +
1.120 + SAMPLETYPE *pMidBuffer;
1.121 + SAMPLETYPE *pMidBufferUnaligned;
1.122 + int overlapLength;
1.123 + int seekLength;
1.124 + int seekWindowLength;
1.125 + int overlapDividerBits;
1.126 + int slopingDivider;
1.127 + float nominalSkip;
1.128 + float skipFract;
1.129 + FIFOSampleBuffer outputBuffer;
1.130 + FIFOSampleBuffer inputBuffer;
1.131 + bool bQuickSeek;
1.132 +
1.133 + int sampleRate;
1.134 + int sequenceMs;
1.135 + int seekWindowMs;
1.136 + int overlapMs;
1.137 + bool bAutoSeqSetting;
1.138 + bool bAutoSeekSetting;
1.139 +
1.140 + void acceptNewOverlapLength(int newOverlapLength);
1.141 +
1.142 + virtual void clearCrossCorrState();
1.143 + void calculateOverlapLength(int overlapMs);
1.144 +
1.145 + virtual double calcCrossCorr(const SAMPLETYPE *mixingPos, const SAMPLETYPE *compare, double &norm) const;
1.146 + virtual double calcCrossCorrAccumulate(const SAMPLETYPE *mixingPos, const SAMPLETYPE *compare, double &norm) const;
1.147 +
1.148 + virtual int seekBestOverlapPositionFull(const SAMPLETYPE *refPos);
1.149 + virtual int seekBestOverlapPositionQuick(const SAMPLETYPE *refPos);
1.150 + int seekBestOverlapPosition(const SAMPLETYPE *refPos);
1.151 +
1.152 + virtual void overlapStereo(SAMPLETYPE *output, const SAMPLETYPE *input) const;
1.153 + virtual void overlapMono(SAMPLETYPE *output, const SAMPLETYPE *input) const;
1.154 + virtual void overlapMulti(SAMPLETYPE *output, const SAMPLETYPE *input) const;
1.155 +
1.156 + void clearMidBuffer();
1.157 + void overlap(SAMPLETYPE *output, const SAMPLETYPE *input, uint ovlPos) const;
1.158 +
1.159 + void calcSeqParameters();
1.160 +
1.161 + /// Changes the tempo of the given sound samples.
1.162 + /// Returns amount of samples returned in the "output" buffer.
1.163 + /// The maximum amount of samples that can be returned at a time is set by
1.164 + /// the 'set_returnBuffer_size' function.
1.165 + void processSamples();
1.166 +
1.167 +public:
1.168 + TDStretch();
1.169 + virtual ~TDStretch();
1.170 +
1.171 + /// Operator 'new' is overloaded so that it automatically creates a suitable instance
1.172 + /// depending on if we've a MMX/SSE/etc-capable CPU available or not.
1.173 + static void *operator new(size_t s);
1.174 +
1.175 + /// Use this function instead of "new" operator to create a new instance of this class.
1.176 + /// This function automatically chooses a correct feature set depending on if the CPU
1.177 + /// supports MMX/SSE/etc extensions.
1.178 + static TDStretch *newInstance();
1.179 +
1.180 + /// Returns the output buffer object
1.181 + FIFOSamplePipe *getOutput() { return &outputBuffer; };
1.182 +
1.183 + /// Returns the input buffer object
1.184 + FIFOSamplePipe *getInput() { return &inputBuffer; };
1.185 +
1.186 + /// Sets new target tempo. Normal tempo = 'SCALE', smaller values represent slower
1.187 + /// tempo, larger faster tempo.
1.188 + void setTempo(float newTempo);
1.189 +
1.190 + /// Returns nonzero if there aren't any samples available for outputting.
1.191 + virtual void clear();
1.192 +
1.193 + /// Clears the input buffer
1.194 + void clearInput();
1.195 +
1.196 + /// Sets the number of channels, 1 = mono, 2 = stereo
1.197 + void setChannels(int numChannels);
1.198 +
1.199 + /// Enables/disables the quick position seeking algorithm. Zero to disable,
1.200 + /// nonzero to enable
1.201 + void enableQuickSeek(bool enable);
1.202 +
1.203 + /// Returns nonzero if the quick seeking algorithm is enabled.
1.204 + bool isQuickSeekEnabled() const;
1.205 +
1.206 + /// Sets routine control parameters. These control are certain time constants
1.207 + /// defining how the sound is stretched to the desired duration.
1.208 + //
1.209 + /// 'sampleRate' = sample rate of the sound
1.210 + /// 'sequenceMS' = one processing sequence length in milliseconds
1.211 + /// 'seekwindowMS' = seeking window length for scanning the best overlapping
1.212 + /// position
1.213 + /// 'overlapMS' = overlapping length
1.214 + void setParameters(int sampleRate, ///< Samplerate of sound being processed (Hz)
1.215 + int sequenceMS = -1, ///< Single processing sequence length (ms)
1.216 + int seekwindowMS = -1, ///< Offset seeking window length (ms)
1.217 + int overlapMS = -1 ///< Sequence overlapping length (ms)
1.218 + );
1.219 +
1.220 + /// Get routine control parameters, see setParameters() function.
1.221 + /// Any of the parameters to this function can be NULL, in such case corresponding parameter
1.222 + /// value isn't returned.
1.223 + void getParameters(int *pSampleRate, int *pSequenceMs, int *pSeekWindowMs, int *pOverlapMs) const;
1.224 +
1.225 + /// Adds 'numsamples' pcs of samples from the 'samples' memory position into
1.226 + /// the input of the object.
1.227 + virtual void putSamples(
1.228 + const SAMPLETYPE *samples, ///< Input sample data
1.229 + uint numSamples ///< Number of samples in 'samples' so that one sample
1.230 + ///< contains both channels if stereo
1.231 + );
1.232 +
1.233 + /// return nominal input sample requirement for triggering a processing batch
1.234 + int getInputSampleReq() const
1.235 + {
1.236 + return (int)(nominalSkip + 0.5);
1.237 + }
1.238 +
1.239 + /// return nominal output sample amount when running a processing batch
1.240 + int getOutputBatchSize() const
1.241 + {
1.242 + return seekWindowLength - overlapLength;
1.243 + }
1.244 +};
1.245 +
1.246 +
1.247 +
1.248 +// Implementation-specific class declarations:
1.249 +
1.250 +#ifdef SOUNDTOUCH_ALLOW_MMX
1.251 + /// Class that implements MMX optimized routines for 16bit integer samples type.
1.252 + class TDStretchMMX : public TDStretch
1.253 + {
1.254 + protected:
1.255 + double calcCrossCorr(const short *mixingPos, const short *compare, double &norm) const;
1.256 + double calcCrossCorrAccumulate(const short *mixingPos, const short *compare, double &norm) const;
1.257 + virtual void overlapStereo(short *output, const short *input) const;
1.258 + virtual void clearCrossCorrState();
1.259 + };
1.260 +#endif /// SOUNDTOUCH_ALLOW_MMX
1.261 +
1.262 +
1.263 +#ifdef SOUNDTOUCH_ALLOW_SSE
1.264 + /// Class that implements SSE optimized routines for floating point samples type.
1.265 + class TDStretchSSE : public TDStretch
1.266 + {
1.267 + protected:
1.268 + double calcCrossCorr(const float *mixingPos, const float *compare, double &norm) const;
1.269 + double calcCrossCorrAccumulate(const float *mixingPos, const float *compare, double &norm) const;
1.270 + };
1.271 +
1.272 +#endif /// SOUNDTOUCH_ALLOW_SSE
1.273 +
1.274 +}
1.275 +#endif /// TDStretch_H
