inpro.pitch
Class PitchTracker

java.lang.Object
  edu.cmu.sphinx.util.props.ConfigurableAdapter
      edu.cmu.sphinx.frontend.BaseDataProcessor
          inpro.pitch.PitchTracker

All Implemented Interfaces:

edu.cmu.sphinx.frontend.DataProcessor, edu.cmu.sphinx.util.props.Configurable

Direct Known Subclasses:

LookAheadPitchTracker

public class PitchTracker
extends edu.cmu.sphinx.frontend.BaseDataProcessor

Data Processor for voicing-decision and pitch tracking. pitch tracking usually consists of several steps: - preprocessing (e. g. low pass filtering) - period candidate generation (score, auto-correlation, cross-correlation, normalized cross-correlation) - candidate refinement (e. g. thresholding of candidates) - voicing decision and final path determination (dynamic programming, ...)

Field Summary

private double	candidateScoreThreshold TODO: IS THIS CORRECT?
protected static boolean	debug
private double	energyThreshold the silence threshold below which we don't do any pitch extraction (this avoids division by zero, and especially reduces processing overhead)
private double	lastBestPitch for tracking, we have to keep track of the previously selected lag
private java.util.List<SignalFeatureListener>	listeners
private int	maxLag
private int	minLag
private static double	NO_LAST_BEST_PITCH
private java.util.concurrent.ArrayBlockingQueue<PitchedDoubleData>	pitchAssignmentQueue pitch can only be assigned to a PitchedDoubleData object a little later, when this data is in the center of the signalBuffer (not the newest data on the right of the buffer) but we don't want to buffer the objects before we output them; we therefore output the PDD objects immediately, keep reference to them in this queue and fill in the pitch values once they have been calculated.
static java.lang.String	PROP_CAND_SCORE_THRESHOLD
static java.lang.String	PROP_LISTENERS
static java.lang.String	PROP_MAX_PITCH_HZ
static java.lang.String	PROP_MIN_PITCH_HZ
private static float	SAMPLING_FREQUENCY
private double[]	signalBuffer

Fields inherited from class edu.cmu.sphinx.util.props.ConfigurableAdapter

logger

Constructor Summary

PitchTracker()

Method Summary

double[]	amdf(double[] signal) calculate score of a given signalBuffer and downsample the signalBuffer on the way.
(package private) static PitchCandidate	bestCandidateSelection(java.util.List<PitchCandidate> candidates) select the candidate with the highest quality (deepest minimum in the lagScoreTrajectory
double[]	cmn(double[] smdsf) cumulative mean normalization
static void	functionalTest(java.lang.String[] args, edu.cmu.sphinx.util.props.ConfigurationManager cm)
edu.cmu.sphinx.frontend.Data	getData()
private static java.util.Queue<java.lang.Double>	getReferencePitch(java.lang.String filename)
static void	main(java.lang.String[] args)
void	newProperties(edu.cmu.sphinx.util.props.PropertySheet ps)
(package private) double	parabolicInterpolation(double alpha, double beta, double gamma)
private java.util.List<PitchCandidate>	qualityThresheldCandidates(double[] lagScoreTrajectory) pick global minima in the lagScoreFunction that are smaller than candidateScoreThreshold as candidates and return the least lag.
private void	signalListeners(PitchCandidate selectedCandidate, java.util.List<PitchCandidate> candidates) signal listeners, and apply values to first element of queue
private static PitchCandidate	simplisticCandidateSelection(java.util.List<PitchCandidate> candidates) select the first candidate (with the highest f0)
double[]	smdsf(double[] signal) calculate the sum magnitude difference square function of a signal.
static void	speedTest(edu.cmu.sphinx.frontend.BaseDataProcessor fe)
(package private) PitchCandidate	trackingCandidateSelection(java.util.List<PitchCandidate> candidates) select lag of the candidate that is closest to the most recently selected lag (which is recorded in lastBestPitch)

Methods inherited from class edu.cmu.sphinx.frontend.BaseDataProcessor

getPredecessor, getTimer, initialize, setPredecessor

Methods inherited from class edu.cmu.sphinx.util.props.ConfigurableAdapter

getName, initLogger, toString

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Field Detail

SAMPLING_FREQUENCY

private static final float SAMPLING_FREQUENCY

See Also:

Constant Field Values

PROP_CAND_SCORE_THRESHOLD

@S4Double(defaultValue=0.15)
public static final java.lang.String PROP_CAND_SCORE_THRESHOLD

See Also:

Constant Field Values

candidateScoreThreshold

private double candidateScoreThreshold

TODO: IS THIS CORRECT? a lower threshold will lead to more (potentially less reliable) candidates

PROP_MIN_PITCH_HZ

@S4Double(defaultValue=66.66666666)
public static final java.lang.String PROP_MIN_PITCH_HZ

See Also:

Constant Field Values

maxLag

private int maxLag

PROP_MAX_PITCH_HZ

@S4Double(defaultValue=500.0)
public static final java.lang.String PROP_MAX_PITCH_HZ

See Also:

Constant Field Values

minLag

private int minLag

energyThreshold

private double energyThreshold

the silence threshold below which we don't do any pitch extraction (this avoids division by zero, and especially reduces processing overhead)

signalBuffer

private double[] signalBuffer

pitchAssignmentQueue

private java.util.concurrent.ArrayBlockingQueue<PitchedDoubleData> pitchAssignmentQueue

pitch can only be assigned to a PitchedDoubleData object a little later, when this data is in the center of the signalBuffer (not the newest data on the right of the buffer) but we don't want to buffer the objects before we output them; we therefore output the PDD objects immediately, keep reference to them in this queue and fill in the pitch values once they have been calculated. The size of the queue (and hence the lag with which pitch becomes available) depends on the size of the signalBuffer, which in turn depends on PROP_MIN_PITCH_HZ.

PROP_LISTENERS

@S4ComponentList(type=SignalFeatureListener.class)
public static final java.lang.String PROP_LISTENERS

See Also:

Constant Field Values

listeners

private java.util.List<SignalFeatureListener> listeners

debug

protected static final boolean debug

See Also:

Constant Field Values

NO_LAST_BEST_PITCH

private static final double NO_LAST_BEST_PITCH

See Also:

Constant Field Values

lastBestPitch

private double lastBestPitch

for tracking, we have to keep track of the previously selected lag

Constructor Detail

PitchTracker

public PitchTracker()

Method Detail

amdf

public double[] amdf(double[] signal)

calculate score of a given signalBuffer and downsample the signalBuffer on the way. NOTE: this is currently not used in the algorithm implemented below.

smdsf

public double[] smdsf(double[] signal)

calculate the sum magnitude difference square function of a signal.

cmn

public double[] cmn(double[] smdsf)

cumulative mean normalization

qualityThresheldCandidates

private java.util.List<PitchCandidate> qualityThresheldCandidates(double[] lagScoreTrajectory)

pick global minima in the lagScoreFunction that are smaller than candidateScoreThreshold as candidates and return the least lag.

Returns:

the least lag of all the candidates considered

trackingCandidateSelection

PitchCandidate trackingCandidateSelection(java.util.List<PitchCandidate> candidates)

select lag of the candidate that is closest to the most recently selected lag (which is recorded in lastBestPitch)

Parameters:

candidates -

Returns:

the selected candidate

simplisticCandidateSelection

private static PitchCandidate simplisticCandidateSelection(java.util.List<PitchCandidate> candidates)

select the first candidate (with the highest f0)

bestCandidateSelection

static PitchCandidate bestCandidateSelection(java.util.List<PitchCandidate> candidates)

select the candidate with the highest quality (deepest minimum in the lagScoreTrajectory

parabolicInterpolation

double parabolicInterpolation(double alpha,
                              double beta,
                              double gamma)

getData

public edu.cmu.sphinx.frontend.Data getData()
                                     throws edu.cmu.sphinx.frontend.DataProcessingException

Specified by:

getData in interface edu.cmu.sphinx.frontend.DataProcessor

Specified by:

getData in class edu.cmu.sphinx.frontend.BaseDataProcessor

Throws:

edu.cmu.sphinx.frontend.DataProcessingException

signalListeners

private void signalListeners(PitchCandidate selectedCandidate,
                             java.util.List<PitchCandidate> candidates)

signal listeners, and apply values to first element of queue

Parameters:

selectedCandidate - or null if no candidate/voiceless

newProperties

public void newProperties(edu.cmu.sphinx.util.props.PropertySheet ps)
                   throws edu.cmu.sphinx.util.props.PropertyException

Specified by:

newProperties in interface edu.cmu.sphinx.util.props.Configurable

Overrides:

newProperties in class edu.cmu.sphinx.util.props.ConfigurableAdapter

Throws:

edu.cmu.sphinx.util.props.PropertyException

getReferencePitch

private static java.util.Queue<java.lang.Double> getReferencePitch(java.lang.String filename)

functionalTest

public static void functionalTest(java.lang.String[] args,
                                  edu.cmu.sphinx.util.props.ConfigurationManager cm)
                           throws edu.cmu.sphinx.util.props.PropertyException,
                                  edu.cmu.sphinx.frontend.DataProcessingException

Throws:

edu.cmu.sphinx.util.props.PropertyException

edu.cmu.sphinx.frontend.DataProcessingException

speedTest

public static void speedTest(edu.cmu.sphinx.frontend.BaseDataProcessor fe)
                      throws edu.cmu.sphinx.frontend.DataProcessingException

Throws:

edu.cmu.sphinx.frontend.DataProcessingException

main

public static void main(java.lang.String[] args)