Joint-GMM voice conversion training using parallel source and target databases
Reference: A. Kain and M. Macon, “Spectral voice conversion for text-to-speech synthesis,” in Proc. of the IEEE ICASSP 1998,
vol. 1, pp. 285-288.
Depending on the parameters it will train GMMs. For example the ouput in this example will be: sourceF_X_targetF_99_10.jgs
→ numTrainingFiles = 99, numComponents = 10 (10 mixes) Input: two directories source and target containing:
/Neutral-Spike-Conversion/source/train_99/*.wav and *.lab /Neutral-Spike-Conversion/target/train_99/*.wav and *.lab In
these directories it will calculate *.lsf, *.ptc, *.ene Output: