marytts.modules.acoustic

Class Model

java.lang.Object

marytts.modules.acoustic.Model

Direct Known Subclasses:

BoundaryModel, CARTModel, HMMModel, SoPModel

public abstract class Model
extends Object

Base class for acoustic modeling; specific Models should extend this and override methods as needed.

Author:

steiner

Field Summary

Fields

Modifier and Type	Field and Description
protected String	applyTo
protected InputStream	dataStream The stream from which we will read our acoustic model.
protected TargetFeatureComputer	featureComputer The producer of feature vectors for the features in predictionFeatureNames as computed by the feature processors in featureManager.
protected FeatureProcessorManager	featureManager The feature processors used for prediction.
protected String	featureName The name of the predicted acoustic feature, if any.
protected String	predictFrom
protected String	predictionFeatureNames The names of the features used for prediction.
protected String	targetAttributeFormat
protected String	targetAttributeName The attribute into which the predicted acoustic feature should be written.
protected String	voiceName The voice with which this model is associated

Constructor Summary

Constructors

Modifier	Constructor and Description
protected	Model(FeatureProcessorManager featureManager, String voiceName, InputStream dataStream, String targetAttributeName, String targetAttributeFormat, String featureName, String predictFrom, String applyTo) Model constructor

Method Summary

Methods

Modifier and Type	Method and Description
void	applyFromTo(List<Element> predictFromElements, List<Element> applyToElements) Apply this Model to a List of Elements, predicting from a different List of Elements
void	applyTo(List<Element> elements) Apply this Model to a List of Elements, predicting from those same Elements
protected abstract float	evaluate(Target target) Evaluate model on a Target to obtain the target value as a float.
String	getApplyTo()
String	getFeatureName()
String	getPredictFrom()
String	getTargetAttributeName()
protected List<Target>	getTargets(List<Element> elements) For a list of PHONE elements, return a list of Targets, where each Target is constructed from the corresponding Element.
String	getVoiceName()
protected void	load() Try to load this model and set the target feature computer appropriately.
protected abstract void	loadData() Load dataFile for this model; only extension classes know how to do this
protected void	setupFeatureComputer()

Methods inherited from class java.lang.Object

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

Field Detail

dataStream

protected InputStream dataStream

The stream from which we will read our acoustic model.

voiceName

protected String voiceName

The voice with which this model is associated

targetAttributeName

protected String targetAttributeName

The attribute into which the predicted acoustic feature should be written.

targetAttributeFormat

protected String targetAttributeFormat

featureName

protected String featureName

The name of the predicted acoustic feature, if any. The feature processor that will be created from this will read the value from targetAttributeName.

featureManager

protected FeatureProcessorManager featureManager

The feature processors used for prediction.

predictionFeatureNames

protected String predictionFeatureNames

The names of the features used for prediction.

featureComputer

protected TargetFeatureComputer featureComputer

The producer of feature vectors for the features in predictionFeatureNames as computed by the feature processors in featureManager.

predictFrom

protected String predictFrom

applyTo

protected String applyTo

Constructor Detail

Model

protected Model(FeatureProcessorManager featureManager,
     String voiceName,
     InputStream dataStream,
     String targetAttributeName,
     String targetAttributeFormat,
     String featureName,
     String predictFrom,
     String applyTo)

Model constructor

Parameters:

featureManager - the feature processor manager used to compute the symbolic features used for prediction

voiceName - name of the voice

dataStream - data file for this Model

targetAttributeName - attribute in MaryXML to predict

targetAttributeFormat - printf-style format String to specify the attribute value, i.e. "%.3f" to round to 3 decimal places; "%s" by default

featureName - name of the custom continuous feature predicted by this model, or null

predictFrom - key of Element Lists from which to predict values; "segments" by default

applyTo - key of Element Lists to which to apply values; "segments" by default

Method Detail

load

protected final void load()
                   throws MaryConfigurationException

Try to load this model and set the target feature computer appropriately. This must be called from the constructor of subclasses, so that the subclass implementation of loadDataFile() is visible.

Throws:

MaryConfigurationException - if the model cannot be set up properly.

loadData

protected abstract void loadData()
                          throws IOException,
                                 MaryConfigurationException

Load dataFile for this model; only extension classes know how to do this

Throws:

IOException - if any files cannot be properly read

MaryConfigurationException - if files can be read but contain problematic content

setupFeatureComputer

protected final void setupFeatureComputer()
                                   throws MaryConfigurationException

Throws:

MaryConfigurationException

applyTo

public void applyTo(List<Element> elements)
             throws MaryConfigurationException

Apply this Model to a List of Elements, predicting from those same Elements

Parameters:

elements - Elements for which to predict the values

Throws:

MaryConfigurationException - if attribute values cannot be predicted because of an invalid voice configuration

applyFromTo

public void applyFromTo(List<Element> predictFromElements,
               List<Element> applyToElements)
                 throws MaryConfigurationException

Apply this Model to a List of Elements, predicting from a different List of Elements

Parameters:

predictFromElements - Elements from which to predict the values

applyToElements - Elements to which to apply the values predicted by this Model

Throws:

MaryConfigurationException - if attribute values cannot be predicted because of an invalid voice configuration

getTargets

protected List<Target> getTargets(List<Element> elements)

For a list of PHONE elements, return a list of Targets, where each Target is constructed from the corresponding Element.

Parameters:

elements - List of Elements

Returns:

List of Targets

evaluate

protected abstract float evaluate(Target target)
                           throws Exception

Evaluate model on a Target to obtain the target value as a float.

Parameters:

target - target

Returns:

target value

Throws:

Exception - if the target value cannot be predicted

getVoiceName

public String getVoiceName()

Returns:

the name of the voice that this model is associated with

getFeatureName

public String getFeatureName()

Returns:

the featureName

getTargetAttributeName

public String getTargetAttributeName()

Returns:

the targetAttributeName

getPredictFrom

public String getPredictFrom()

Returns:

the key of Element Lists from which to predict with this Model

getApplyTo

public String getApplyTo()

Returns:

the key of Element Lists to which to apply this Model