[mary-users] New voice creation

[Aleś Bułojčyk]

Hi Nickolay.

I tried to recompile voice with new frequency. I didn't resampled input
database to 16kHz. Instead, I set 44100 frequency for output:

    HMMVoiceConfigure.freqWarp = 0.53
    HMMVoiceConfigure.sampfreq = 44100
    HMMVoiceConfigure.frameLen = 1102.5
    HMMVoiceConfigure.frameShift = 220.5

I used http://www.dfki.de/pipermail/mary-users/2012-April/001189.html for
define freqWarp, frameLen and frameShift parameters.

Result is interesting. Looks like quality is much better, but voice sounds
3 times faster. When I make it 3 times slower in audacity, it sounds better
than my previous 16kHz voice.

Could you answer on two my questions:
1. Where is good description for HMMVoiceConfigure parameters ? Looks like
I should change they into some more reasonable values.
2. I send my step-by-step voice creation log in the previous mail. Did I
miss some step ? Is voice training a fully automatic process ? Does voice
training means some manual activity, like listen some intermediate result
and correct some things in the input files ?

WBR, Alex.
​
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.dfki.de/pipermail/mary-users/attachments/20151108/2f152cc2/attachment-0001.htm