Shout_MakeTrainSet Class Reference

This class is the base class for the shout_maketrainset application. It creates a new AM trainset given a Master Label File and an output directory. More...

Inheritance diagram for Shout_MakeTrainSet:

[legend]

List of all members.


Public Member Functions
	Shout_MakeTrainSet (char task, char phoneSetFileName, char *trainDirName)
	Shout_MakeTrainSet (char task, char phoneSetFile, char datFile, char rawName, char rttmFileName, char trainDirName, char decisionTree, char method, char trainList, char speakerFilter, float useP)
	~Shout_MakeTrainSet ()
void	closePhoneFiles (bool printOutput)
void	setFeaturePool (FeaturePool *fp)
void	storePhone (int cdmID, int contextLeft, int phoneID, int contextRight, int b1, int b2, int b3, int numberOfObservations)
Protected Member Functions
void	openPhoneFiles ()
void	checkDataDir (bool allowCDM)
Protected Attributes
bool	performCluster
TASK_TYPE	performWhat
FILE **	phoneFile
DataStats *	data
int	nrOfModels
int	nrOfDecisionRules
int	nrOfDecisionRuleSamples [MAX_NUMBER_OF_DECISION_RULES][MAX_CLUSTERS]
int	nrOfSil
char *	trainDirName
FeaturePool *	featurePool
int	totalPhonesStored
int	vectorSize
int	skippedLines
float	useProbability

Detailed Description

This class is the base class for the shout_maketrainset application. It creates a new AM trainset given a Master Label File and an output directory.

Constructor & Destructor Documentation

Shout_MakeTrainSet::Shout_MakeTrainSet	(	char *	task,
		char *	phoneSetFileName,
		char *	trainDirName
	)

References checkDataDir(), data, nrOfModels, DataStats::nrOfSamples, nrOfSil, openPhoneFiles(), performCluster, performWhat, skippedLines, TASK_TYPE_DIAR, TASK_TYPE_HIST_NORM, TASK_TYPE_PHONE, totalPhonesStored, trainDirName, and vectorSize.

Here is the call graph for this function:

Shout_MakeTrainSet::Shout_MakeTrainSet	(	char *	task,
		char *	phoneSetFileName,
		char *	datFile,
		char *	rawName,
		char *	rttmFileName,
		char *	tD,
		char *	decisionTree,
		char *	method,
		char *	trainList,
		char *	speakerFilter,
		float	useP
	)

This constructor will read a phone definition file and a hypothesis file. If both are consistent, it will write the features of the phones from the MLF to the output training directory. ShoutTrainMaster reads the data from this file during training.

Todo:: I'm in a hurry because our harddrive is full with accumulator data! Make sure to clean up this bit of code later!

References Vector::addElements(), checkDataDir(), FeaturePool::claimSegmentationID(), FeaturePool::createNewPool(), data, featurePool, FILE_TYPE_RAW_DECODING, FILE_TYPE_RAW_DIARIZATION, FILE_TYPE_RAW_HISTOGRAM_NORM, FILE_TYPE_RAW_SAD, FILE_TYPE_RAW_SPKREC, FILE_TYPE_RAW_VTLN, WriteFileLittleBigEndian::freadEndianSafe(), FeaturePool::getVectorSize(), nrOfDecisionRules, nrOfDecisionRuleSamples, nrOfModels, nrOfSil, openPhoneFiles(), performCluster, performWhat, FeaturePool::readRTTM(), Vector::setAllValues(), skippedLines, Vector::storeData(), storePhone(), TASK_TYPE_ARTICULATORY, TASK_TYPE_DIAR, TASK_TYPE_HIST_NORM, TASK_TYPE_PHONE, TASK_TYPE_SAD, TASK_TYPE_UNDEF, TASK_TYPE_VTLN, totalPhonesStored, trainDirName, useProbability, and vectorSize.

Here is the call graph for this function:

Shout_MakeTrainSet::~Shout_MakeTrainSet ( )

The destructor gets rid of some administration...

References closePhoneFiles(), skippedLines, and trainDirName.

Here is the call graph for this function:

Member Function Documentation

void Shout_MakeTrainSet::checkDataDir ( bool allowCDM ) [protected]

Checks if the acoustic data in the training directory is valid. If allowCDM is false, a warning is printed when the data contains Cluster Dependent Model information.

References data, WriteFileLittleBigEndian::freadEndianSafe(), DataStats::name, nrOfDecisionRules, nrOfDecisionRuleSamples, nrOfModels, DataStats::nrOfSamples, nrOfSil, performWhat, TASK_TYPE_ARTICULATORY, TASK_TYPE_DIAR, TASK_TYPE_HIST_NORM, TASK_TYPE_PHONE, trainDirName, and vectorSize.

Referenced by Shout_MakeTrainSet().

Here is the call graph for this function:

void Shout_MakeTrainSet::closePhoneFiles ( bool printOutput )

Writes final statistics are written to disc and if printOutput is true, some statistics are written to standard output.

References data, nrOfDecisionRules, nrOfDecisionRuleSamples, nrOfModels, DataStats::nrOfSamples, nrOfSil, performWhat, phoneFile, TASK_TYPE_ARTICULATORY, TASK_TYPE_HIST_NORM, TASK_TYPE_SAD, TASK_TYPE_VTLN, totalPhonesStored, trainDirName, and vectorSize.

Referenced by ~Shout_MakeTrainSet().

void Shout_MakeTrainSet::openPhoneFiles ( ) [protected]

Creates a file handle to each phone data file.

References data, nrOfDecisionRules, nrOfModels, performWhat, phoneFile, TASK_TYPE_ARTICULATORY, TASK_TYPE_HIST_NORM, TASK_TYPE_SAD, TASK_TYPE_VTLN, and trainDirName.

Referenced by Shout_MakeTrainSet().

void Shout_MakeTrainSet::setFeaturePool ( FeaturePool * fp )

Sets a new feature pool for storing phones..

References featurePool.

void Shout_MakeTrainSet::storePhone	(	int	cdmID,
		int	contextLeft,
		int	phoneID,
		int	contextRight,
		int	b1,
		int	b2,
		int	b3,
		int	numberOfObservations
	)

Stores a phone to the data directory.

cdmID determines for which speaker cluster this data is used.
contextLeft and contextRight determine which phones are at the left- and right of this phone. (not always used)
phoneID is the ID of the phone itself.
bl, b2, b3 and e3 provide state alignment information.

References data, featurePool, Vector::getValue(), FeaturePool::getVector(), DataStats::nrOfSamples, nrOfSil, performWhat, phoneFile, Vector::storeData(), TASK_TYPE_ARTICULATORY, totalPhonesStored, and useProbability.