Weighted codebook based voice conversion algorithms.
The core weighted codebook mapping algorithm (WeightedCodebookParallelTrainer)
is extended to enable frame based mapping as well within the same class.
Classical codebook mapping produces more smooth results since the source and target acoustic mapping is
done using average spectral feature vectors corresponding to each source and target phoneme pair observed in the training data.
Frame mapping goes one step beyond to directly map the source and target frame-level features.
Therefore, it is expected to result in more detail with an increased probability of output discontinuities.
This can be compensated by to some extent using the temporal transformation function smoother, marytts.signalproc.adaptation.smoothing.TemporalSmoother