GMM based estimator (for fast VTLN, GID, etc.) More...

#include <bsapi.h>

Inheritance diagram for SGMMBasedEstimatorI:

Public Member Functions
virtual void BSAPI_METHOD	SetModelSuffix (const char *pModelSuffix="gmm")=0
	Set the model file suffix.
virtual bool BSAPI_METHOD	LoadModels (const char *pModelDirectory)=0
	Load models files given by dictionary and file suffix.
virtual void BSAPI_METHOD	SetTarget (SParametersCallbackI *pTarget)=0
	Set target object for receiving updated factor and scores.
virtual bool BSAPI_METHOD	SetEstimationMode (const char *pEstimationMode="sum")=0
	Set the estimation mode.
virtual void BSAPI_METHOD	SetSpeechLabel (const char *pSpeechLabel="voice")=0
	Set speech label to be used with segmentation.
virtual void BSAPI_METHOD	SetFrameShiftMs (int frameShiftMs=10)=0
	Set frame shift.
virtual void BSAPI_METHOD	SetUseSegmentation (bool useSegmentation=true)=0
	Set whether the segmentation will be applied after the `OnFeatureMatrix` has been called.
virtual void BSAPI_METHOD	SetBufferLength (int bufferLength=0)=0
virtual void BSAPI_METHOD	SetFrameBunchLength (int frameBunchLength=50)=0
virtual void BSAPI_METHOD	SetAdaptMinFrames (int adaptMinFrames=50)=0
virtual void BSAPI_METHOD	SetAdaptTimeConst (float adaptTimeConst=0.02f)=0
virtual void BSAPI_METHOD	SetOutputStep (int outputStep=0)=0
virtual void BSAPI_METHOD	SetUseFactor (bool useFactor=false)=0
virtual void BSAPI_METHOD	SetDefaultFactor (float defaultFactor=1.0f)=0
virtual bool BSAPI_METHOD	SetFileNameFactorRange (const char *pRange="1:3")=0
virtual void BSAPI_METHOD	SetFileNameFactorScale (float scale=0.01f)=0
virtual void BSAPI_METHOD	SetFrameStride (int frameStride=1)=0
virtual bool BSAPI_METHOD	AddFeatures (SFloatMatrixI pFeatures, int nFrames, segm_pos_type pSegments=0, int nSegments=0, unsigned int flags=PF_LASTFRAME\|PF_OFFLINEPROC)=0
	Add features for estimation.
virtual int BSAPI_METHOD	GetNUsedFrames ()=0
	Get number of frames used for estimation.
virtual float BSAPI_METHOD	GetFactor ()=0
	Get actual estimate of the factor.
virtual float *BSAPI_METHOD	GetClassScores (int *pNumValues=0)=0
	Get scores of all classes.
virtual char **BSAPI_METHOD	GetClassNames (int *pNumValues=0)=0
	Get names of all classes.
virtual char *BSAPI_METHOD	GetBestClass (float *pScore=0)=0
	Get the name of the winning class.
virtual bool BSAPI_METHOD	Restart ()=0
	Restart estimation.

Detailed Description

GMM based estimator (for fast VTLN, GID, etc.)

Member Function Documentation

virtual bool BSAPI_METHOD SGMMBasedEstimatorI::AddFeatures	(	SFloatMatrixI *	pFeatures,
		int	nFrames,
		segm_pos_type *	pSegments = `0`,
		int	nSegments = `0`,
		unsigned int	flags = `PF_LASTFRAME\|PF_OFFLINEPROC`
	)		`[pure virtual]`

Add features for estimation.

Adds features to compute new scores and factor estimation. Segmentation can be optionally given too to tell which segments (feature vectors) should be used in estimation. Those segments must be labeled SE_VOICE and their boundaries must be in HTK format (1 unit is equal to 100 nanoseconds). Note that if internal segmentation is enabled (specified in configuration file) it is used no matter what segmentation was passed to this method.

Parameters:

pFeatures	feature vectors to be add for estimation
nFrames	number of feature vectors
pSegments	pointer to array with segmentation or 0 to use all feature vectors
nSegments	length of the segmentation array
flags	processing flags

See also:: AddWaveform; SFeatureExtractionCallbackI; STransCallbackI

virtual char* BSAPI_METHOD SGMMBasedEstimatorI::GetBestClass ( float * pScore = 0 ) [pure virtual]

Get the name of the winning class.

Parameters:

pScore pointer where to store winning class score

Returns:: winning class name

virtual char** BSAPI_METHOD SGMMBasedEstimatorI::GetClassNames ( int * pNumValues = 0 ) [pure virtual]

Get names of all classes.

Parameters:

pNumValues pointer where to store names array length

Returns:: class names array

virtual float* BSAPI_METHOD SGMMBasedEstimatorI::GetClassScores ( int * pNumValues = 0 ) [pure virtual]

Get scores of all classes.

Parameters:

pNumValues pointer where to store scores array length

Returns:: class scores array

virtual float BSAPI_METHOD SGMMBasedEstimatorI::GetFactor ( ) [pure virtual]

Get actual estimate of the factor.

Returns:: factor

virtual int BSAPI_METHOD SGMMBasedEstimatorI::GetNUsedFrames ( ) [pure virtual]

Get number of frames used for estimation.

Returns:: number of frames

virtual bool BSAPI_METHOD SGMMBasedEstimatorI::LoadModels ( const char * pModelDirectory ) [pure virtual]

Load models files given by dictionary and file suffix.

Returns:: true it the models were loaded correctly.

virtual bool BSAPI_METHOD SGMMBasedEstimatorI::Restart ( ) [pure virtual]

Restart estimation.

Flushes all internal buffers and resets object to its initial state (factor would be set to a default value).

Returns:: true restart finished correctly

virtual bool BSAPI_METHOD SGMMBasedEstimatorI::SetEstimationMode ( const char * pEstimationMode = "sum" ) [pure virtual]

Set the estimation mode.

Parameters:

pEstimationMode estimation mode - 'sum' for averiging, 'adapt' for online adaptation (resent values are more important).

Returns:: false if the mode string is unknown.

virtual void BSAPI_METHOD SGMMBasedEstimatorI::SetFrameShiftMs ( int frameShiftMs = 10 ) [pure virtual]

Set frame shift.

Sets the frame shift in milliseconds. Frame shift is used only with AddFeatures and OnFeatureMatrix. Otherwise it is obtained automatically from parameterization configuration. Correct frame shift is necessary for use with segmentation however it is also needed for correct values of time parameter passed to OnParameters callback.

Parameters:

frameShiftMs frame shift in milliseconds

virtual void BSAPI_METHOD SGMMBasedEstimatorI::SetModelSuffix ( const char * pModelSuffix = "gmm" ) [pure virtual]

Set the model file suffix.

Parameters:

pModelSuffix model suffix. 'gmm' by default.

virtual void BSAPI_METHOD SGMMBasedEstimatorI::SetSpeechLabel ( const char * pSpeechLabel = "voice" ) [pure virtual]

Set speech label to be used with segmentation.

Sets the label value so the segments with this label coming to the OnTranscription callback will be considered as speech and will be processed

Parameters:

pSpeechLabel label of (speech) segments to be processed

See also:: SParametersCallbackI

virtual void BSAPI_METHOD SGMMBasedEstimatorI::SetTarget ( SParametersCallbackI * pTarget ) [pure virtual]

Set target object for receiving updated factor and scores.

Parameters:

pTarget pointer to target object

virtual void BSAPI_METHOD SGMMBasedEstimatorI::SetUseSegmentation ( bool useSegmentation = true ) [pure virtual]

Set whether the segmentation will be applied after the OnFeatureMatrix has been called.

Tells whether after OnFeatureMatrix callback also OnTranscription will be called to pass segmentation or not. Using this method has no effect to the way how AddFeatures works.

Parameters:

useSegmentation

true await segmentation will be passed after OnFeatureMatrix by OnTranscription
false segmentation is not used, OnTranscription is not expected to be called

See also:: SFeatureExtractionCallbackI; STransCallbackI

The documentation for this class was generated from the following file:

bsapi.h

Public Member Functions

Detailed Description

Member Function Documentation