Package marytts.signalproc.adaptation

Packages for speaking style and speaker identity adaptation in Mary TTS supporting various voice conversion algorithms.
The subpackages include:
.codebook: Weighted codebook and weighted frame mapping
.gmm: Gaussian mixture model based spectral conversion
    .gmm.jointgmm: Joint source-target GMM based voice conversion (this method has been used in our group�s various publications in 2008 and 2009)

See:
          Description

Class Summary
AdaptationUtils Generic utilities for voice conversion
BaselineAdaptationItem This class keeps information on each specific training item For example, a training item for a sentence based voice conversion training database could be a wav file, the corresponding text transcription, label file, pitch contour file, etc.
BaselineAdaptationSet Baseline class for speaker specific voice conversion training set
BaselineFeatureCollection Baseline class for indexed binary files of acoustic feature sets
BaselineFeatureExtractor Baseline class for acoustic feature analysis for voice conversion
BaselineMappingFileHeader Baseline class for voice conversion function binary file header
BaselineParams Baseline voice conversion parameters.
BaselinePostprocessor Baseline class for post-processing operations after voice conversion
BaselinePreprocessor Baseline class for pre-processing operations after voice conversion
BaselineTrainer Baseline class for voice conversion training
BaselineTrainerParams  
BaselineTransformer Baseline class for voice conversion transformation
BaselineTransformerParams Baseline class for voice conversion transformation parameters All specific implementations of transformation stage of a given voice conversion algorithm should use a parameter set that is derived from this class
Context A wrapper class for representing phonetic context
FdpsolaAdapter A class that supports voice conversion through weighted codebook mapping or joint-GMMs and FDPSOLA based prosody and vocal tract modifications Reference: Moulines, E.
FileCopier Generic utility class for renaming and copying voice conversion training files
FileMap A class for handling source-target frame indices mapping for a single pair
IndexMap A class for handling source-target frame indices mapping It can be used in various ways:
LsfMatch  
MfccAdapter This class transforms MFCCs to MFCCs
OutlierStatus Flags for outliers in source and target training data
TargetLsfCopyMapper  
VocalTractMatch  
VocalTractTransformationData Baseline class for voice conversion transformation data
VocalTractTransformationFunction Baseline class for vocal tract transformation functions for voice conversion
 

Package marytts.signalproc.adaptation Description

Packages for speaking style and speaker identity adaptation in Mary TTS supporting various voice conversion algorithms.
The subpackages include:
.codebook: Weighted codebook and weighted frame mapping
.gmm: Gaussian mixture model based spectral conversion
    .gmm.jointgmm: Joint source-target GMM based voice conversion (this method has been used in our group�s various publications in 2008 and 2009)


IMPORTANT NOTE:
For expressive speech transformation, GMM based technology works better as compared to codebook and frame mapping based methods.
For speaker identity conversion, opposite results are observed in various papers.