marytts.modules.acoustic

Class HMMModel

java.lang.Object

marytts.modules.acoustic.Model

marytts.modules.acoustic.HMMModel

public class HMMModel
extends Model

Model for predicting duration and F0 from HMMs

Author:

marcela

Field Summary

Fields

Modifier and Type	Field and Description
protected static org.apache.log4j.Logger	logger

Fields inherited from class marytts.modules.acoustic.Model

applyTo, dataStream, featureComputer, featureManager, featureName, predictFrom, predictionFeatureNames, targetAttributeFormat, targetAttributeName, voiceName

Constructor Summary

Constructors

Constructor and Description
HMMModel(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) Predict F0 for the list of elements and apply to another list of elements.
void	applyTo(List<Element> elements) Predict duration for the list of elements.
protected float	evaluate(Target target) Apply the HMM to a Target to get its predicted value, this method is not used in HMMModel.
protected void	loadData() Load trees and pdfs, from HMM configuration file.
void	setPredictDurAndF0(boolean bval) This variable is set to true whenever the same HMMModel is used to predict both duration and F0.

Methods inherited from class marytts.modules.acoustic.Model

getApplyTo, getFeatureName, getPredictFrom, getTargetAttributeName, getTargets, getVoiceName, load, setupFeatureComputer

Methods inherited from class java.lang.Object

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

Field Detail

logger

protected static org.apache.log4j.Logger logger

Constructor Detail

HMMModel

public HMMModel(FeatureProcessorManager featureManager,
        String voiceName,
        InputStream dataStream,
        String targetAttributeName,
        String targetAttributeFormat,
        String featureName,
        String predictFrom,
        String applyTo)
         throws MaryConfigurationException

Model constructor

Parameters:

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

voiceName - in HMM models this data file corresponds to the configuration file of the HMM voice

dataStream - dataStream

targetAttributeName - attribute in MARYXML to predict

targetAttributeFormat - print style, not used in HMM models

featureName - not used in HMMModel

predictFrom - not used in HMMModel

applyTo - not used in HMMModel

Throws:

MaryConfigurationException - if there are missing files or problems loading trees and pdf files.

Method Detail

setPredictDurAndF0

public void setPredictDurAndF0(boolean bval)

This variable is set to true whenever the same HMMModel is used to predict both duration and F0. by default the variable is false, so that means that two different HMMModels are used for predicting duration and F0, in this case there is no state durations information to predict F0.

Parameters:

bval - bval

loadData

protected void loadData()
                 throws IOException,
                        MaryConfigurationException

Load trees and pdfs, from HMM configuration file.

Specified by:

loadData in class Model

Throws:

MaryConfigurationException - if there are missing files or problems loading trees and pdf files.

IOException - if any files cannot be properly read

applyTo

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

Predict duration for the list of elements. If the same HMMModel is used to predict duration and F0 then a utterance model is created and kept in a WeakHashMap, so the next call to this module, for predicting F0, can use that utterance model.

Overrides:

applyTo in class Model

Parameters:

elements - elements from MaryXML for which to predict the values

Throws:

MaryConfigurationException - if error searching in HMM trees.

applyFromTo

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

Predict F0 for the list of elements and apply to another list of elements. If the same HMMModel is used to predict duration and F0 then there must be a utterance model created in a previous call to this module, that will be used to predict F0. If there is no previously created utterance model then one is created.

Overrides:

applyFromTo in class Model

Parameters:

predictFromElements - elements from MaryXML for which to predict the values

applyToElements - elements from MaryXML for which to apply the predicted values

Throws:

MaryConfigurationException - if error searching in HMM trees.

evaluate

protected float evaluate(Target target)

Apply the HMM to a Target to get its predicted value, this method is not used in HMMModel.

Specified by:

evaluate in class Model

Parameters:

target - target

Returns:

target value

Throws:

RuntimeException - if this method is called.