Prosody transformation algorithms for voice conversion.
A prosody modification framework has been implemented which supports:
Mean and standard deviation transformation of f0
Sentence slope transformation
Mean and standard deviation transformation is the best method so far.
Duration and energy transformation have not yet been implemented.
A collection of analysis algorithms for signal processing.
Important classes are as follows:
LpcAnalyser: Linear prediction analysis using autocorrelation
appraoch and Durbin recursion
LsfAnalyser: Computation of line spectral frequencies (LSFs, or
line spectral pairs - LSPs) based on LpcAnalyser
EnergyAnalyser: Energy contour estimation with voice activity
F0TrackerAutocorrelationHeuristic: An autocorrelation based
f0 analysis algorithm extended with heuristic post-processing to reduce
voiced/unvoiced errors and f0 doubling/halving problems.