edu.cmu.sphinx.decoder.search

Class WordPruningBreadthFirstLookaheadSearchManager

java.lang.Object

edu.cmu.sphinx.decoder.search.TokenSearchManager

edu.cmu.sphinx.decoder.search.WordPruningBreadthFirstSearchManager

edu.cmu.sphinx.decoder.search.WordPruningBreadthFirstLookaheadSearchManager

All Implemented Interfaces:

SearchManager, Configurable

public class WordPruningBreadthFirstLookaheadSearchManager
extends WordPruningBreadthFirstSearchManager

Provides the breadth first search with fast match heuristic included to reduce amount of tokens created.

All scores and probabilities are maintained in the log math log domain.

Field Summary

Fields

Modifier and Type	Field and Description
protected ActiveList	fastmatchActiveList
protected java.util.Map<SearchState,Token>	fastMatchBestTokenMap
static java.lang.String	PROP_FASTMATCH_LINGUIST The property that defines the name of the linguist to be used for fast match.
static java.lang.String	PROP_FM_ACTIVE_LIST_FACTORY
static java.lang.String	PROP_LOADER The property that to get direct access to gau for score caching control.
static java.lang.String	PROP_LOOKAHEAD_PENALTY_WEIGHT
static java.lang.String	PROP_LOOKAHEAD_WINDOW The property that controls size of lookahead window.

Fields inherited from class edu.cmu.sphinx.decoder.search.WordPruningBreadthFirstSearchManager

acousticLookaheadFrames, activeList, bestTokenMap, checkStateOrder, currentCollectTime, currentFrameNumber, curTokensScored, growTimer, linguist, logger, logMath, loserManager, PROP_ACOUSTIC_LOOKAHEAD_FRAMES, PROP_ACTIVE_LIST_MANAGER, PROP_CHECK_STATE_ORDER, PROP_GROW_SKIP_INTERVAL, PROP_LINGUIST, PROP_MAX_LATTICE_EDGES, PROP_PRUNER, PROP_RELATIVE_BEAM_WIDTH, PROP_SCORER, PROP_SHOW_TOKEN_COUNT, pruner, pruneTimer, relativeBeamWidth, resultList, scorer, scoreTimer, showTokenCount, streamEnd, tokensCreated, totalTokensScored

Fields inherited from class edu.cmu.sphinx.decoder.search.TokenSearchManager

buildWordLattice, keepAllTokens, PROP_BUILD_WORD_LATTICE, PROP_KEEP_ALL_TOKENS

Constructor Summary

Constructors

Constructor and Description
WordPruningBreadthFirstLookaheadSearchManager()
WordPruningBreadthFirstLookaheadSearchManager(Linguist linguist, Linguist fastmatchLinguist, Loader loader, Pruner pruner, AcousticScorer scorer, ActiveListManager activeListManager, ActiveListFactory fastmatchActiveListFactory, boolean showTokenCount, double relativeWordBeamWidth, int growSkipInterval, boolean checkStateOrder, boolean buildWordLattice, int lookaheadWindow, float lookaheadWeight, int maxLatticeEdges, float acousticLookaheadFrames, boolean keepAllTokens) Creates a pruning manager with lookahead

Method Summary

Modifier and Type	Method and Description
protected void	collectFastMatchSuccessorTokens(Token token)
protected void	collectSuccessorTokens(Token token) Collects the next set of emitting tokens from a token and accumulates them in the active or result lists
protected void	createFastMatchBestTokenMap() creates a new best token map with the best size
protected Token	getFastMatchBestToken(SearchState state)
protected void	growFastmatchBranches() Goes through the fast match active list of tokens and expands each token, finding the set of successor tokens until all the successor tokens are emitting tokens.
protected void	localStart() Gets the initial grammar node from the linguist and creates a GrammarNodeToken
void	newProperties(PropertySheet ps) This method is called when this configurable component needs to be reconfigured.
protected void	pruneFastMatchBranches() Removes unpromising branches from the fast match active list
Result	recognize(int nFrames) Performs the recognition for the given number of frames.
protected boolean	scoreFastMatchTokens()
protected void	setFastMatchBestToken(Token token, SearchState state)

Methods inherited from class edu.cmu.sphinx.decoder.search.WordPruningBreadthFirstSearchManager

activeListAdd, allocate, allowExpansion, checkStateOrder, createBestTokenMap, deallocate, getActiveList, getBestToken, getCurrentFrameNumber, getGrowTimer, getResultList, getTokensCreated, growBranches, growEmittingBranches, isVisited, localStop, monitorStates, pruneBranches, recognize, scoreTokens, setActiveList, setBestToken, setResultList, showTokenCount, startRecognition, stopRecognition

Methods inherited from class edu.cmu.sphinx.decoder.search.TokenSearchManager

getResultListPredecessor

Methods inherited from class java.lang.Object

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

Field Detail

PROP_LOADER

@S4Component(type=Loader.class)
public static final java.lang.String PROP_LOADER

The property that to get direct access to gau for score caching control.

See Also:

Constant Field Values

PROP_FASTMATCH_LINGUIST

@S4Component(type=Linguist.class)
public static final java.lang.String PROP_FASTMATCH_LINGUIST

The property that defines the name of the linguist to be used for fast match.

See Also:

Constant Field Values

PROP_FM_ACTIVE_LIST_FACTORY

@S4Component(type=ActiveListFactory.class)
public static final java.lang.String PROP_FM_ACTIVE_LIST_FACTORY

See Also:

Constant Field Values

PROP_LOOKAHEAD_PENALTY_WEIGHT

@S4Double(defaultValue=1.0)
public static final java.lang.String PROP_LOOKAHEAD_PENALTY_WEIGHT

See Also:

Constant Field Values

PROP_LOOKAHEAD_WINDOW

@S4Integer(defaultValue=5)
public static final java.lang.String PROP_LOOKAHEAD_WINDOW

The property that controls size of lookahead window. Acceptable values are in range [1..10].

See Also:

Constant Field Values

fastmatchActiveList

protected ActiveList fastmatchActiveList

fastMatchBestTokenMap

protected java.util.Map<SearchState,Token> fastMatchBestTokenMap

Constructor Detail

WordPruningBreadthFirstLookaheadSearchManager

public WordPruningBreadthFirstLookaheadSearchManager(Linguist linguist,
                                                     Linguist fastmatchLinguist,
                                                     Loader loader,
                                                     Pruner pruner,
                                                     AcousticScorer scorer,
                                                     ActiveListManager activeListManager,
                                                     ActiveListFactory fastmatchActiveListFactory,
                                                     boolean showTokenCount,
                                                     double relativeWordBeamWidth,
                                                     int growSkipInterval,
                                                     boolean checkStateOrder,
                                                     boolean buildWordLattice,
                                                     int lookaheadWindow,
                                                     float lookaheadWeight,
                                                     int maxLatticeEdges,
                                                     float acousticLookaheadFrames,
                                                     boolean keepAllTokens)

Creates a pruning manager with lookahead

Parameters:

linguist - a linguist for search space

fastmatchLinguist - a linguist for fast search space

pruner - pruner to drop tokens

loader - model loader

scorer - scorer to estimate token probability

activeListManager - active list manager to store tokens

fastmatchActiveListFactory - fast match active list factor to store phoneloop tokens

showTokenCount - show count during decoding

relativeWordBeamWidth - relative beam for lookahead pruning

growSkipInterval - skip interval for grown

checkStateOrder - check order of states during growth

buildWordLattice - build a lattice during decoding

maxLatticeEdges - max edges to keep in lattice

acousticLookaheadFrames - frames to do lookahead

keepAllTokens - keep tokens including emitting tokens

lookaheadWindow - window for lookahead

lookaheadWeight - weight for lookahead pruning

WordPruningBreadthFirstLookaheadSearchManager

public WordPruningBreadthFirstLookaheadSearchManager()

Method Detail

newProperties

public void newProperties(PropertySheet ps)
                   throws PropertyException

Description copied from interface: Configurable

This method is called when this configurable component needs to be reconfigured.

Specified by:

newProperties in interface Configurable

Overrides:

newProperties in class WordPruningBreadthFirstSearchManager

Parameters:

ps - a property sheet holding the new data

Throws:

PropertyException - if there is a problem with the properties.

recognize

public Result recognize(int nFrames)

Performs the recognition for the given number of frames.

Specified by:

recognize in interface SearchManager

Overrides:

recognize in class WordPruningBreadthFirstSearchManager

Parameters:

nFrames - the number of frames to recognize

Returns:

the current result

createFastMatchBestTokenMap

protected void createFastMatchBestTokenMap()

creates a new best token map with the best size

localStart

protected void localStart()

Gets the initial grammar node from the linguist and creates a GrammarNodeToken

Overrides:

localStart in class WordPruningBreadthFirstSearchManager

growFastmatchBranches

protected void growFastmatchBranches()

Goes through the fast match active list of tokens and expands each token, finding the set of successor tokens until all the successor tokens are emitting tokens.

scoreFastMatchTokens

protected boolean scoreFastMatchTokens()

pruneFastMatchBranches

protected void pruneFastMatchBranches()

Removes unpromising branches from the fast match active list

getFastMatchBestToken

protected Token getFastMatchBestToken(SearchState state)

setFastMatchBestToken

protected void setFastMatchBestToken(Token token,
                                     SearchState state)

collectFastMatchSuccessorTokens

protected void collectFastMatchSuccessorTokens(Token token)

collectSuccessorTokens

protected void collectSuccessorTokens(Token token)

Collects the next set of emitting tokens from a token and accumulates them in the active or result lists

Overrides:

collectSuccessorTokens in class WordPruningBreadthFirstSearchManager

Parameters:

token - the token to collect successors from be immediately expanded are placed. Null if we should always expand all nodes.