Chris@1: /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*-  vi:set ts=8 sts=4 sw=4: */
Chris@1: 
Chris@1: /*
Chris@1:     Vamp feature extraction plugin for the BeatRoot beat tracker.
Chris@1: 
Chris@1:     Centre for Digital Music, Queen Mary, University of London.
Chris@1:     This file copyright 2011 Simon Dixon, Chris Cannam and QMUL.
Chris@1:     
Chris@1:     This program is free software; you can redistribute it and/or
Chris@1:     modify it under the terms of the GNU General Public License as
Chris@1:     published by the Free Software Foundation; either version 2 of the
Chris@1:     License, or (at your option) any later version.  See the file
Chris@1:     COPYING included with this distribution for more information.
Chris@1: */
Chris@1: 
Chris@1: #ifndef _BEATROOT_PROCESSOR_H_
Chris@1: #define _BEATROOT_PROCESSOR_H_
Chris@1: 
Chris@1: class BeatRootProcessor
Chris@1: {
Chris@1: protected:
Chris@1:     /** Sample rate of audio */
Chris@1:     float sampleRate;
Chris@1: 	
Chris@1:     /** Spacing of audio frames (determines the amount of overlap or
Chris@1:      *  skip between frames). This value is expressed in
Chris@1:      *  seconds. (Default = 0.020s) */
Chris@1:     double hopTime;
Chris@1: 
Chris@1:     /** The approximate size of an FFT frame in seconds. (Default =
Chris@1:      *  0.04644s).  The value is adjusted so that <code>fftSize</code>
Chris@1:      *  is always power of 2. */
Chris@1:     double fftTime;
Chris@1: 
Chris@1:     /** Spacing of audio frames in samples (see <code>hopTime</code>) */
Chris@1:     int hopSize;
Chris@1: 
Chris@1:     /** The size of an FFT frame in samples (see <code>fftTime</code>) */
Chris@1:     int fftSize;
Chris@1: 
Chris@1:     /** The number of overlapping frames of audio data which have been read. */
Chris@1:     int frameCount;
Chris@1: 
Chris@1:     /** RMS amplitude of the current frame. */
Chris@1:     double frameRMS;
Chris@1: 
Chris@1:     /** Long term average frame energy (in frequency domain representation). */
Chris@1:     double ltAverage;
Chris@1: 
Chris@1:     /** Spectral flux onset detection function, indexed by frame. */
Chris@1:     vector<int> spectralFlux;
Chris@1: 	
Chris@1:     /** A mapping function for mapping FFT bins to final frequency bins.
Chris@1:      *  The mapping is linear (1-1) until the resolution reaches 2 points per
Chris@1:      *  semitone, then logarithmic with a semitone resolution.  e.g. for
Chris@1:      *  44.1kHz sampling rate and fftSize of 2048 (46ms), bin spacing is
Chris@1:      *  21.5Hz, which is mapped linearly for bins 0-34 (0 to 732Hz), and
Chris@1:      *  logarithmically for the remaining bins (midi notes 79 to 127, bins 35 to
Chris@1:      *  83), where all energy above note 127 is mapped into the final bin. */
Chris@1:     vector<int> freqMap;
Chris@1: 
Chris@1:     /** The number of entries in <code>freqMap</code>. Note that the length of
Chris@1:      *  the array is greater, because its size is not known at creation time. */
Chris@1:     int freqMapSize;
Chris@1: 
Chris@1:     /** The magnitude spectrum of the most recent frame.  Used for
Chris@1:      *  calculating the spectral flux. */
Chris@1:     vector<double> prevFrame;
Chris@1: 	
Chris@1:     /** The magnitude spectrum of the current frame. */
Chris@1:     vector<double> newFrame;
Chris@1: 
Chris@1:     /** The magnitude spectra of all frames, used for plotting the spectrogram. */
Chris@1:     vector<vector<double> > frames; //!!! do we need this? much cheaper to lose it if we don't
Chris@1: 	
Chris@1:     /** The RMS energy of all frames. */
Chris@1:     vector<double> energy; //!!! unused in beat tracking?
Chris@1: 	
Chris@1:     /** The estimated onset times from peak-picking the onset
Chris@1:      * detection function(s). */
Chris@1:     vector<double> onsets;
Chris@1: 	
Chris@1:     /** The estimated onset times and their saliences. */	
Chris@1:     //!!!EventList onsetList;
Chris@1:     vector<double> onsetList; //!!! corresponding to keyDown member of events in list
Chris@1: 
Chris@1:     /** Total number of audio frames if known, or -1 for live or compressed input. */
Chris@1:     int totalFrames;
Chris@1: 	
Chris@1:     /** Flag for enabling or disabling debugging output */
Chris@1:     static bool debug = false;
Chris@1: 	
Chris@1:     /** Flag for suppressing all standard output messages except results. */
Chris@1:     static bool silent = true;
Chris@1: 	
Chris@1:     /** RMS frame energy below this value results in the frame being
Chris@1:      *  set to zero, so that normalisation does not have undesired
Chris@1:      *  side-effects. */
Chris@1:     static double silenceThreshold = 0.0004; //!!!??? energy of what? should not be static?
Chris@1: 	
Chris@1:     /** For dynamic range compression, this value is added to the log
Chris@1:      *  magnitude in each frequency bin and any remaining negative
Chris@1:      *  values are then set to zero.
Chris@1:      */
Chris@1:     static double rangeThreshold = 10; //!!! sim
Chris@1: 	
Chris@1:     /** Determines method of normalisation. Values can be:<ul>
Chris@1:      *  <li>0: no normalisation</li>
Chris@1:      *  <li>1: normalisation by current frame energy</li>
Chris@1:      *  <li>2: normalisation by exponential average of frame energy</li>
Chris@1:      *  </ul>
Chris@1:      */
Chris@1:     static int normaliseMode = 2;
Chris@1: 	
Chris@1:     /** Ratio between rate of sampling the signal energy (for the
Chris@1:      * amplitude envelope) and the hop size */
Chris@1:     static int energyOversampleFactor = 2; //!!! not used?
Chris@1: 	
Chris@1: public:
Chris@1: 
Chris@1:     /** Constructor: note that streams are not opened until the input
Chris@1:      *  file is set (see <code>setInputFile()</code>). */
Chris@1:     AudioProcessor() {
Chris@1:         cbIndex = 0;
Chris@1:         frameRMS = 0;
Chris@1:         ltAverage = 0;
Chris@1:         frameCount = 0;
Chris@1:         hopSize = 0;
Chris@1:         fftSize = 0;
Chris@1:         hopTime = 0.010;	// DEFAULT, overridden with -h
Chris@1:         fftTime = 0.04644;	// DEFAULT, overridden with -f
Chris@1:         progressCallback = null;
Chris@1:         stdIn = new BufferedReader(new InputStreamReader(System.in));
Chris@1:         if (doOnsetPlot)
Chris@1:             plot = new Plot();
Chris@1:     } // constructor
Chris@1: 
Chris@1: 	/** For debugging, outputs information about the AudioProcessor to
Chris@1: 	 *  standard error.
Chris@1: 	 */
Chris@1: 	public void print() {
Chris@1: 		System.err.println(this);
Chris@1: 	} // print()
Chris@1: 
Chris@1: 	/** For interactive pause - wait for user to hit Enter */
Chris@1: 	public String readLine() {
Chris@1: 		try { return stdIn.readLine(); } catch (Exception e) { return null; }
Chris@1: 	} // readLine()
Chris@1: 	
Chris@1: 	/** Gives some basic information about the audio being processed. */
Chris@1: 	public String toString() {
Chris@1: 		return "AudioProcessor\n" +
Chris@1: 				String.format("\tFile: %s (%3.1f kHz, %1d channels)\n",
Chris@1: 						audioFileName, sampleRate/1000, channels) +
Chris@1: 				String.format("\tHop / FFT sizes: %5.3f / %5.3f",
Chris@1: 						hopTime, hopTime * fftSize / hopSize);
Chris@1: 	} // toString()
Chris@1: 
Chris@1: 	/** Adds a link to the GUI component which shows the progress of matching.
Chris@1: 	 *  @param c the AudioProcessor representing the other performance 
Chris@1: 	 */
Chris@1: 	public void setProgressCallback(ProgressIndicator c) {
Chris@1: 		progressCallback = c;
Chris@1: 	} // setProgressCallback()
Chris@1: 
Chris@1: 	/** Sets up the streams and buffers for live audio input (CD quality).
Chris@1: 	 *  If any Exception is thrown within this method, it is caught, and any
Chris@1: 	 *  opened streams are closed, and <code>pcmInputStream</code> is set to
Chris@1: 	 *  <code>null</code>, indicating that the method did not complete
Chris@1: 	 *  successfully.
Chris@1: 	 */
Chris@1: 	public void setLiveInput() {
Chris@1: 		try {
Chris@1: 			channels = 2;
Chris@1: 			sampleRate = 44100;
Chris@1: 			AudioFormat desiredFormat = new AudioFormat(
Chris@1: 						AudioFormat.Encoding.PCM_SIGNED, sampleRate, 16,
Chris@1: 						channels, channels * 2, sampleRate, false);
Chris@1: 			TargetDataLine tdl = AudioSystem.getTargetDataLine(desiredFormat);
Chris@1: 			tdl.open(desiredFormat, liveInputBufferSize);
Chris@1: 			pcmInputStream = new AudioInputStream(tdl);
Chris@1: 			audioFormat = pcmInputStream.getFormat();
Chris@1: 			init();
Chris@1: 			tdl.start();
Chris@1: 		} catch (Exception e) {
Chris@1: 			e.printStackTrace();
Chris@1: 			closeStreams();	// make sure it exits in a consistent state
Chris@1: 		}
Chris@1: 	} // setLiveInput()
Chris@1: 
Chris@1: 	/** Sets up the streams and buffers for audio file input.
Chris@1: 	 *  If any Exception is thrown within this method, it is caught, and any
Chris@1: 	 *  opened streams are closed, and <code>pcmInputStream</code> is set to
Chris@1: 	 *  <code>null</code>, indicating that the method did not complete
Chris@1: 	 *  successfully.
Chris@1: 	 *  @param fileName The path name of the input audio file.
Chris@1: 	 */
Chris@1: 	public void setInputFile(String fileName) {
Chris@1: 		closeStreams();		// release previously allocated resources
Chris@1: 		audioFileName = fileName;
Chris@1: 		try {
Chris@1: 			if (audioFileName == null)
Chris@1: 				throw new Exception("No input file specified");
Chris@1: 			File audioFile = new File(audioFileName);
Chris@1: 			if (!audioFile.isFile())
Chris@1: 				throw new FileNotFoundException(
Chris@1: 							"Requested file does not exist: " + audioFileName);
Chris@1: 			rawInputStream = AudioSystem.getAudioInputStream(audioFile);
Chris@1: 			audioFormat = rawInputStream.getFormat();
Chris@1: 			channels = audioFormat.getChannels();
Chris@1: 			sampleRate = audioFormat.getSampleRate();
Chris@1: 			pcmInputStream = rawInputStream;
Chris@1: 			if ((audioFormat.getEncoding()!=AudioFormat.Encoding.PCM_SIGNED) ||
Chris@1: 					(audioFormat.getFrameSize() != channels * 2) ||
Chris@1: 					audioFormat.isBigEndian()) {
Chris@1: 				AudioFormat desiredFormat = new AudioFormat(
Chris@1: 						AudioFormat.Encoding.PCM_SIGNED, sampleRate, 16,
Chris@1: 						channels, channels * 2, sampleRate, false);
Chris@1: 				pcmInputStream = AudioSystem.getAudioInputStream(desiredFormat,
Chris@1: 																rawInputStream);
Chris@1: 				audioFormat = desiredFormat;
Chris@1: 			}
Chris@1: 			init();
Chris@1: 		} catch (Exception e) {
Chris@1: 			e.printStackTrace();
Chris@1: 			closeStreams();	// make sure it exits in a consistent state
Chris@1: 		}
Chris@1: 	} // setInputFile()
Chris@1: 
Chris@1: 	/** Allocates memory for arrays, based on parameter settings */
Chris@1: 	protected void init() {
Chris@1: 		hopSize = (int) Math.round(sampleRate * hopTime);
Chris@1: 		fftSize = (int) Math.round(Math.pow(2,
Chris@1: 				Math.round( Math.log(fftTime * sampleRate) / Math.log(2))));
Chris@1: 		makeFreqMap(fftSize, sampleRate);
Chris@1: 		int buffSize = hopSize * channels * 2;
Chris@1: 		if ((inputBuffer == null) || (inputBuffer.length != buffSize))
Chris@1: 			inputBuffer = new byte[buffSize];
Chris@1: 		if ((circBuffer == null) || (circBuffer.length != fftSize)) {
Chris@1: 			circBuffer = new double[fftSize];
Chris@1: 			reBuffer = new double[fftSize];
Chris@1: 			imBuffer = new double[fftSize];
Chris@1: 			prevPhase = new double[fftSize];
Chris@1: 			prevPrevPhase = new double[fftSize];
Chris@1: 			prevFrame = new double[fftSize];
Chris@1: 			window = FFT.makeWindow(FFT.HAMMING, fftSize, fftSize);
Chris@1: 			for (int i=0; i < fftSize; i++)
Chris@1: 				window[i] *= Math.sqrt(fftSize);
Chris@1: 		}
Chris@1: 		if (pcmInputStream == rawInputStream)
Chris@1: 			totalFrames = (int)(pcmInputStream.getFrameLength() / hopSize);
Chris@1: 		else
Chris@1: 			totalFrames = (int) (MAX_LENGTH / hopTime);
Chris@1: 		if ((newFrame == null) || (newFrame.length != freqMapSize)) {
Chris@1: 			newFrame = new double[freqMapSize];
Chris@1: 			frames = new double[totalFrames][freqMapSize];
Chris@1: 		} else if (frames.length != totalFrames)
Chris@1: 			frames = new double[totalFrames][freqMapSize];
Chris@1: 		energy = new double[totalFrames*energyOversampleFactor];
Chris@1: 		phaseDeviation = new double[totalFrames];
Chris@1: 		spectralFlux = new double[totalFrames];
Chris@1: 		frameCount = 0;
Chris@1: 		cbIndex = 0;
Chris@1: 		frameRMS = 0;
Chris@1: 		ltAverage = 0;
Chris@1: 		progressCallback = null;
Chris@1: 	} // init()
Chris@1: 
Chris@1: 	/** Closes the input stream(s) associated with this object. */
Chris@1: 	public void closeStreams() {
Chris@1: 		if (pcmInputStream != null) {
Chris@1: 			try {
Chris@1: 				pcmInputStream.close();
Chris@1: 				if (pcmInputStream != rawInputStream)
Chris@1: 					rawInputStream.close();
Chris@1: 				if (audioOut != null) {
Chris@1: 					audioOut.drain();
Chris@1: 					audioOut.close();
Chris@1: 				}
Chris@1: 			} catch (Exception e) {}
Chris@1: 			pcmInputStream = null;
Chris@1: 			audioOut = null;
Chris@1: 		}
Chris@1: 	} // closeStreams()
Chris@1: 
Chris@1: 	/** Creates a map of FFT frequency bins to comparison bins.
Chris@1: 	 *  Where the spacing of FFT bins is less than 0.5 semitones, the mapping is
Chris@1: 	 *  one to one. Where the spacing is greater than 0.5 semitones, the FFT
Chris@1: 	 *  energy is mapped into semitone-wide bins. No scaling is performed; that
Chris@1: 	 *  is the energy is summed into the comparison bins. See also
Chris@1: 	 *  processFrame()
Chris@1: 	 */
Chris@1: 	protected void makeFreqMap(int fftSize, float sampleRate) {
Chris@1: 		freqMap = new int[fftSize/2+1];
Chris@1: 		double binWidth = sampleRate / fftSize;
Chris@1: 		int crossoverBin = (int)(2 / (Math.pow(2, 1/12.0) - 1));
Chris@1: 		int crossoverMidi = (int)Math.round(Math.log(crossoverBin*binWidth/440)/
Chris@1: 														Math.log(2) * 12 + 69);
Chris@1: 		// freq = 440 * Math.pow(2, (midi-69)/12.0) / binWidth;
Chris@1: 		int i = 0;
Chris@1: 		while (i <= crossoverBin)
Chris@1: 			freqMap[i++] = i;
Chris@1: 		while (i <= fftSize/2) {
Chris@1: 			double midi = Math.log(i*binWidth/440) / Math.log(2) * 12 + 69;
Chris@1: 			if (midi > 127)
Chris@1: 				midi = 127;
Chris@1: 			freqMap[i++] = crossoverBin + (int)Math.round(midi) - crossoverMidi;
Chris@1: 		}
Chris@1: 		freqMapSize = freqMap[i-1] + 1;
Chris@1: 	} // makeFreqMap()
Chris@1: 
Chris@1: 	/** Calculates the weighted phase deviation onset detection function.
Chris@1: 	 *  Not used.
Chris@1: 	 *  TODO: Test the change to WPD fn */
Chris@1: 	protected void weightedPhaseDeviation() {
Chris@1: 		if (frameCount < 2)
Chris@1: 			phaseDeviation[frameCount] = 0;
Chris@1: 		else {
Chris@1: 			for (int i = 0; i < fftSize; i++) {
Chris@1: 				double pd = imBuffer[i] - 2 * prevPhase[i] + prevPrevPhase[i];
Chris@1: 				double pd1 = Math.abs(Math.IEEEremainder(pd, 2 * Math.PI));
Chris@1: 				phaseDeviation[frameCount] += pd1 * reBuffer[i];
Chris@1: 				// System.err.printf("%7.3f   %7.3f\n", pd/Math.PI, pd1/Math.PI);
Chris@1: 			}
Chris@1: 		}
Chris@1: 		phaseDeviation[frameCount] /= fftSize * Math.PI;
Chris@1: 		double[] tmp = prevPrevPhase;
Chris@1: 		prevPrevPhase = prevPhase;
Chris@1: 		prevPhase = imBuffer;
Chris@1: 		imBuffer = tmp;
Chris@1: 	} // weightedPhaseDeviation()
Chris@1: 
Chris@1: 	/** Reads a frame of input data, averages the channels to mono, scales
Chris@1: 	 *  to a maximum possible absolute value of 1, and stores the audio data
Chris@1: 	 *  in a circular input buffer.
Chris@1: 	 *  @return true if a frame (or part of a frame, if it is the final frame)
Chris@1: 	 *  is read. If a complete frame cannot be read, the InputStream is set
Chris@1: 	 *  to null.
Chris@1: 	 */
Chris@1: 	public boolean getFrame() {
Chris@1: 		if (pcmInputStream == null)
Chris@1: 			return false;
Chris@1: 		try {
Chris@1: 			int bytesRead = (int) pcmInputStream.read(inputBuffer);
Chris@1: 			if ((audioOut != null) && (bytesRead > 0))
Chris@1: 				if (audioOut.write(inputBuffer, 0, bytesRead) != bytesRead)
Chris@1: 					System.err.println("Error writing to audio device");
Chris@1: 			if (bytesRead < inputBuffer.length) {
Chris@1: 				if (!silent)
Chris@1: 					System.err.println("End of input: " + audioFileName);
Chris@1: 				closeStreams();
Chris@1: 				return false;
Chris@1: 			}
Chris@1: 		} catch (IOException e) {
Chris@1: 			e.printStackTrace();
Chris@1: 			closeStreams();
Chris@1: 			return false;
Chris@1: 		}
Chris@1: 		frameRMS = 0;
Chris@1: 		double sample;
Chris@1: 		switch(channels) {
Chris@1: 			case 1:
Chris@1: 				for (int i = 0; i < inputBuffer.length; i += 2) {
Chris@1: 					sample = ((inputBuffer[i+1]<<8) |
Chris@1: 							  (inputBuffer[i]&0xff)) / 32768.0;
Chris@1: 					frameRMS += sample * sample;
Chris@1: 					circBuffer[cbIndex++] = sample;
Chris@1: 					if (cbIndex == fftSize)
Chris@1: 						cbIndex = 0;
Chris@1: 				}
Chris@1: 				break;
Chris@1: 			case 2: // saves ~0.1% of RT (total input overhead ~0.4%) :)
Chris@1: 				for (int i = 0; i < inputBuffer.length; i += 4) {
Chris@1: 					sample = (((inputBuffer[i+1]<<8) | (inputBuffer[i]&0xff)) +
Chris@1: 							  ((inputBuffer[i+3]<<8) | (inputBuffer[i+2]&0xff)))
Chris@1: 								/ 65536.0;
Chris@1: 					frameRMS += sample * sample;
Chris@1: 					circBuffer[cbIndex++] = sample;
Chris@1: 					if (cbIndex == fftSize)
Chris@1: 						cbIndex = 0;
Chris@1: 				}
Chris@1: 				break;
Chris@1: 			default:
Chris@1: 				for (int i = 0; i < inputBuffer.length; ) {
Chris@1: 					sample = 0;
Chris@1: 					for (int j = 0; j < channels; j++, i+=2)
Chris@1: 						sample += (inputBuffer[i+1]<<8) | (inputBuffer[i]&0xff);
Chris@1: 					sample /= 32768.0 * channels;
Chris@1: 					frameRMS += sample * sample;
Chris@1: 					circBuffer[cbIndex++] = sample;
Chris@1: 					if (cbIndex == fftSize)
Chris@1: 						cbIndex = 0;
Chris@1: 				}
Chris@1: 		}
Chris@1: 		frameRMS = Math.sqrt(frameRMS / inputBuffer.length * 2 * channels);
Chris@1: 		return true;
Chris@1: 	} // getFrame()
Chris@1: 
Chris@1: 	/** Processes a frame of audio data by first computing the STFT with a
Chris@1: 	 *  Hamming window, then mapping the frequency bins into a part-linear
Chris@1: 	 *  part-logarithmic array, then computing the spectral flux 
Chris@1: 	 *  then (optionally) normalising and calculating onsets.
Chris@1: 	 */
Chris@1: 	protected void processFrame() {
Chris@1: 		if (getFrame()) {
Chris@1: 			for (int i = 0; i < fftSize; i++) {
Chris@1: 				reBuffer[i] = window[i] * circBuffer[cbIndex];
Chris@1: 				if (++cbIndex == fftSize)
Chris@1: 					cbIndex = 0;
Chris@1: 			}
Chris@1: 			Arrays.fill(imBuffer, 0);
Chris@1: 			FFT.magnitudePhaseFFT(reBuffer, imBuffer);
Chris@1: 			Arrays.fill(newFrame, 0);
Chris@1: 			double flux = 0;
Chris@1: 			for (int i = 0; i <= fftSize/2; i++) {
Chris@1: 				if (reBuffer[i] > prevFrame[i])
Chris@1: 					flux += reBuffer[i] - prevFrame[i];
Chris@1: 				newFrame[freqMap[i]] += reBuffer[i];
Chris@1: 			}
Chris@1: 			spectralFlux[frameCount] = flux;
Chris@1: 			for (int i = 0; i < freqMapSize; i++)
Chris@1: 				frames[frameCount][i] = newFrame[i];
Chris@1: 			int index = cbIndex - (fftSize - hopSize);
Chris@1: 			if (index < 0)
Chris@1: 				index += fftSize;
Chris@1: 			int sz = (fftSize - hopSize) / energyOversampleFactor;
Chris@1: 			for (int j = 0; j < energyOversampleFactor; j++) {
Chris@1: 				double newEnergy = 0;
Chris@1: 				for (int i = 0; i < sz; i++) {
Chris@1: 					newEnergy += circBuffer[index] * circBuffer[index];
Chris@1: 					if (++index == fftSize)
Chris@1: 						index = 0;
Chris@1: 				}
Chris@1: 				energy[frameCount * energyOversampleFactor + j] =
Chris@1: 						newEnergy / sz <= 1e-6? 0: Math.log(newEnergy / sz) + 13.816;
Chris@1: 			}
Chris@1: 			double decay = frameCount >= 200? 0.99:
Chris@1: 						(frameCount < 100? 0: (frameCount - 100) / 100.0);
Chris@1: 			if (ltAverage == 0)
Chris@1: 				ltAverage = frameRMS;
Chris@1: 			else
Chris@1: 				ltAverage = ltAverage * decay + frameRMS * (1.0 - decay);
Chris@1: 			if (frameRMS <= silenceThreshold)
Chris@1: 				for (int i = 0; i < freqMapSize; i++)
Chris@1: 					frames[frameCount][i] = 0;
Chris@1: 			else {
Chris@1: 				if (normaliseMode == 1)
Chris@1: 					for (int i = 0; i < freqMapSize; i++)
Chris@1: 						frames[frameCount][i] /= frameRMS;
Chris@1: 				else if (normaliseMode == 2)
Chris@1: 					for (int i = 0; i < freqMapSize; i++)
Chris@1: 						frames[frameCount][i] /= ltAverage;
Chris@1: 				for (int i = 0; i < freqMapSize; i++) {
Chris@1: 					frames[frameCount][i] = Math.log(frames[frameCount][i]) + rangeThreshold;
Chris@1: 					if (frames[frameCount][i] < 0)
Chris@1: 						frames[frameCount][i] = 0;
Chris@1: 				}
Chris@1: 			}
Chris@1: //			weightedPhaseDeviation();
Chris@1: //			if (debug)
Chris@1: //				System.err.printf("PhaseDev:  t=%7.3f  phDev=%7.3f  RMS=%7.3f\n",
Chris@1: //						frameCount * hopTime,
Chris@1: //						phaseDeviation[frameCount],
Chris@1: //						frameRMS);
Chris@1: 			double[] tmp = prevFrame;
Chris@1: 			prevFrame = reBuffer;
Chris@1: 			reBuffer = tmp;
Chris@1: 			frameCount++;
Chris@1: 			if ((frameCount % 100) == 0) {
Chris@1: 				if (!silent) {
Chris@1: 					System.err.printf("Progress: %1d %5.3f %5.3f\n", 
Chris@1: 							frameCount, frameRMS, ltAverage);
Chris@1: 					Profile.report();
Chris@1: 				}
Chris@1: 				if ((progressCallback != null) && (totalFrames > 0))
Chris@1: 					progressCallback.setFraction((double)frameCount/totalFrames);
Chris@1: 			}
Chris@1: 		}
Chris@1: 	} // processFrame()
Chris@1: 
Chris@1: 	/** Processes a complete file of audio data. */
Chris@1: 	public void processFile() {
Chris@1: 		while (pcmInputStream != null) {
Chris@1: 			// Profile.start(0);
Chris@1: 			processFrame();
Chris@1: 			// Profile.log(0);
Chris@1: 			if (Thread.currentThread().isInterrupted()) {
Chris@1: 				System.err.println("info: INTERRUPTED in processFile()");
Chris@1: 				return;
Chris@1: 			}
Chris@1: 		}
Chris@1: 
Chris@1: //		double[] x1 = new double[phaseDeviation.length];
Chris@1: //		for (int i = 0; i < x1.length; i++) {
Chris@1: //			x1[i] = i * hopTime;
Chris@1: //			phaseDeviation[i] = (phaseDeviation[i] - 0.4) * 100;
Chris@1: //		}
Chris@1: //		double[] x2 = new double[energy.length];
Chris@1: //		for (int i = 0; i < x2.length; i++)
Chris@1: //			x2[i] = i * hopTime / energyOversampleFactor;
Chris@1: //		// plot.clear();
Chris@1: //		plot.addPlot(x1, phaseDeviation, Color.green, 7);
Chris@1: //		plot.addPlot(x2, energy, Color.red, 7);
Chris@1: //		plot.setTitle("Test phase deviation");
Chris@1: //		plot.fitAxes();
Chris@1: 
Chris@1: //		double[] slope = new double[energy.length];
Chris@1: //		double hop = hopTime / energyOversampleFactor;
Chris@1: //		Peaks.getSlope(energy, hop, 15, slope);
Chris@1: //		LinkedList<Integer> peaks = Peaks.findPeaks(slope, (int)Math.round(0.06 / hop), 10);
Chris@1: 		
Chris@1: 		double hop = hopTime;
Chris@1: 		Peaks.normalise(spectralFlux);
Chris@1: 		LinkedList<Integer> peaks = Peaks.findPeaks(spectralFlux, (int)Math.round(0.06 / hop), 0.35, 0.84, true);
Chris@1: 		onsets = new double[peaks.size()];
Chris@1: 		double[] y2 = new double[onsets.length];
Chris@1: 		Iterator<Integer> it = peaks.iterator();
Chris@1: 		onsetList = new EventList();
Chris@1: 		double minSalience = Peaks.min(spectralFlux);
Chris@1: 		for (int i = 0; i < onsets.length; i++) {
Chris@1: 			int index = it.next();
Chris@1: 			onsets[i] = index * hop;
Chris@1: 			y2[i] = spectralFlux[index];
Chris@1: 			Event e = BeatTrackDisplay.newBeat(onsets[i], 0);
Chris@1: //			if (debug)
Chris@1: //				System.err.printf("Onset: %8.3f  %8.3f  %8.3f\n",
Chris@1: //						onsets[i], energy[index], slope[index]);
Chris@1: //			e.salience = slope[index];	// or combination of energy + slope??
Chris@1: 			// Note that salience must be non-negative or the beat tracking system fails!
Chris@1: 			e.salience = spectralFlux[index] - minSalience;
Chris@1: 			onsetList.add(e);
Chris@1: 		}
Chris@1: 		if (progressCallback != null)
Chris@1: 			progressCallback.setFraction(1.0);
Chris@1: 		if (doOnsetPlot) {
Chris@1: 			double[] x1 = new double[spectralFlux.length];
Chris@1: 			for (int i = 0; i < x1.length; i++)
Chris@1: 				x1[i] = i * hopTime;
Chris@1: 			plot.addPlot(x1, spectralFlux, Color.red, 4);
Chris@1: 			plot.addPlot(onsets, y2, Color.green, 3);
Chris@1: 			plot.setTitle("Spectral flux and onsets");
Chris@1: 			plot.fitAxes();
Chris@1: 		}
Chris@1: 		if (debug) {
Chris@1: 			System.err.printf("Onsets: %d\nContinue? ", onsets.length);
Chris@1: 			readLine();
Chris@1: 		}
Chris@1: 	} // processFile()
Chris@1: 
Chris@1: 	/** Reads a text file containing a list of whitespace-separated feature values.
Chris@1: 	 *  Created for paper submitted to ICASSP'07.
Chris@1: 	 *  @param fileName File containing the data
Chris@1: 	 *  @return An array containing the feature values
Chris@1: 	 */
Chris@1: 	public static double[] getFeatures(String fileName) {
Chris@1: 		ArrayList<Double> l = new ArrayList<Double>();
Chris@1: 		try {
Chris@1: 			BufferedReader b = new BufferedReader(new FileReader(fileName));
Chris@1: 			while (true) {
Chris@1: 				String s = b.readLine();
Chris@1: 				if (s == null)
Chris@1: 					break;
Chris@1: 				int start = 0;
Chris@1: 				while (start < s.length()) {
Chris@1: 					int len = s.substring(start).indexOf(' ');
Chris@1: 					String t = null;
Chris@1: 					if (len < 0)
Chris@1: 						t = s.substring(start);
Chris@1: 					else if (len > 0) {
Chris@1: 						t = s.substring(start, start + len);
Chris@1: 					}
Chris@1: 					if (t != null)
Chris@1: 						try {
Chris@1: 							l.add(Double.parseDouble(t));
Chris@1: 						} catch (NumberFormatException e) {
Chris@1: 							System.err.println(e);
Chris@1: 							if (l.size() == 0)
Chris@1: 								l.add(new Double(0));
Chris@1: 							else
Chris@1: 								l.add(new Double(l.get(l.size()-1)));
Chris@1: 						}
Chris@1: 					start += len + 1;
Chris@1: 					if (len < 0)
Chris@1: 						break;
Chris@1: 				}
Chris@1: 			}
Chris@1: 			double[] features = new double[l.size()];
Chris@1: 			Iterator<Double> it = l.iterator();
Chris@1: 			for (int i = 0; it.hasNext(); i++)
Chris@1: 				features[i] = it.next().doubleValue();
Chris@1: 			return features;
Chris@1: 		} catch (FileNotFoundException e) {
Chris@1: 			e.printStackTrace();
Chris@1: 			return null;
Chris@1: 		} catch (IOException e) {
Chris@1: 			e.printStackTrace();
Chris@1: 			return null;
Chris@1: 		} catch (NumberFormatException e) {
Chris@1: 			e.printStackTrace();
Chris@1: 			return null;
Chris@1: 		}
Chris@1: 	} // getFeatures()
Chris@1: 	
Chris@1: 	/** Reads a file of feature values, treated as an onset detection function,
Chris@1: 	 *  and finds peaks, which are stored in <code>onsetList</code> and <code>onsets</code>.
Chris@1: 	 * @param fileName The file of feature values
Chris@1: 	 * @param hopTime The spacing of feature values in time
Chris@1: 	 */
Chris@1: 	public void processFeatures(String fileName, double hopTime) {
Chris@1: 		double hop = hopTime;
Chris@1: 		double[] features = getFeatures(fileName);
Chris@1: 		Peaks.normalise(features);
Chris@1: 		LinkedList<Integer> peaks = Peaks.findPeaks(features, (int)Math.round(0.06 / hop), 0.35, 0.84, true);
Chris@1: 		onsets = new double[peaks.size()];
Chris@1: 		double[] y2 = new double[onsets.length];
Chris@1: 		Iterator<Integer> it = peaks.iterator();
Chris@1: 		onsetList = new EventList();
Chris@1: 		double minSalience = Peaks.min(features);
Chris@1: 		for (int i = 0; i < onsets.length; i++) {
Chris@1: 			int index = it.next();
Chris@1: 			onsets[i] = index * hop;
Chris@1: 			y2[i] = features[index];
Chris@1: 			Event e = BeatTrackDisplay.newBeat(onsets[i], 0);
Chris@1: 			e.salience = features[index] - minSalience;
Chris@1: 			onsetList.add(e);
Chris@1: 		}
Chris@1: 	} // processFeatures()
Chris@1: 
Chris@1: 	/** Copies output of audio processing to the display panel. */
Chris@1: 	public void setDisplay(BeatTrackDisplay btd) {
Chris@1: 		int energy2[] = new int[totalFrames*energyOversampleFactor];
Chris@1: 		double time[] = new double[totalFrames*energyOversampleFactor];
Chris@1: 		for (int i = 0; i < totalFrames*energyOversampleFactor; i++) {
Chris@1: 			energy2[i] = (int) (energy[i] * 4 * energyOversampleFactor);
Chris@1: 			time[i] = i * hopTime / energyOversampleFactor;
Chris@1: 		}
Chris@1: 		btd.setMagnitudes(energy2);
Chris@1: 		btd.setEnvTimes(time);
Chris@1: 		btd.setSpectro(frames, totalFrames, hopTime, 0);//fftTime/hopTime);
Chris@1: 		btd.setOnsets(onsets);
Chris@1: 		btd.setOnsetList(onsetList);
Chris@1: 	} // setDisplay()
Chris@1: 	
Chris@1: } // class AudioProcessor
Chris@1: 
Chris@1: 
Chris@1: #endif