Antonio Bonafonte wrote:
Dear all, We have created several hts voices. And the male voices, sound 'whispering'. I have changed the -u parameter in hts_engine, and in some cases it improves a little .. but not enough. I have checked just a few .lf0 files and they seem ok. (But maybe we should check more files).
Hi Antonio,yes - check more F0 files. I had a similar problem and it was due to the pitch tracker incorrectly marking too many voiced sections as unvoiced in the training data. As a result, the trained model tended to generate unvoiced frames, leading to a hoarse or whispered voice.
Try tweaking your pitch tracker, or use a different one. Simon