[hts-users:01949] Re: speaker adaptation help

[王翔 <wangxiang68@xxxxxxxxx>]

Yes, I can synthesize voice using average model.

But using xforms to do HMMGens I just synthesize silence or other noises.
The whole processing was running just with some warnings like "HERest can't find *** in the edge...", probably means the data of two speakers are not match?

Thanks

2009/4/12 Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>

Hi,

王翔 wrote:

hi
I have run speaker adaptation module of HTS successfuly, speaker 1 and 2 are women, use 100 sentence to do MLLR transformation.
1) But the synsthesized voice by HMMGens and sptk was silence.
   Can you tell me why?

There is little information.
Can you synthesize voice using average model?
First, you should check the voice synthesized by using average model.

2) what is the meaning of MLLR Block and Bandwidth.  If i use the 41*3 dimention mgc, 5*3 dimention  lf0, and 8 dimention duration, how should i configure these parameters?

This is transformation matrix.
You should refer to this.

C.J. Leggetter and P.C. Woodland, “Maximum likelihood linear
regression for speaker adaptation of continuous density hidden
Markov models,” Computer Speech & Language, vol. 9, pp.
171–185, 1995.

Regards,
Keiichiro Oura