[hts-users:02050] Re: Problem Feature Extraction

["Heiga ZEN (Byung Ha CHUN)" <heiga.zen@xxxxxxxxxxxxxxxxx>]

Hi,

wangxiang wrote:

> Yes, I use dmp to print out the coefficients. The attachment is the mgc-lsp.
>
> $(X2X) +af $${sp} | \
>
> $(MGCEP) -a $(FREQWARP) -g $(GAMMA) -m $(MGCORDER) -l 1024 -f 1e-20000 
> -q 3 -o 4 | \
>
> $(LPC2LSP) -m $(MGCORDER) -s $${SAMPKHZ} $${GAINOPT} -n 1024 -p 8 -d 
> 1e-8 > mgc/$${base}.mgc; \
>
> this time the coefficient seems right, and the training process is 
> smooth, but using hts-engine it synthesizes silence ...

I guess you're using spectra extracted by MATLAB STRAIGHT for $${sp}.  MATLAB 
STRAIGHT internally normalizes dynamic range of input waveform while extracting 
spectra.  This is the reason why dynamic range of synthesized waveform is 
re-normalized in HTS-demo/data/scripts/synthesis.m as

[sy] = exstraightsynth(f0,sp,ap,16000,prm);
wavwrite( sy/max(abs(sy)), 16000, 'temp.wav');

hts_engine doesn't have such 're-normalization' functionality.  If amplitude of 
synthesized waveform (without re-normalization) ranges within [-1,...,+1], the 
final output becomes complete silence.

Regards,

Heiga ZEN (Byung Ha CHUN)

--
--------------------------
Heiga ZEN (Byung Ha CHUN)
Speech Technology Group
Cambridge Research Lab
Toshiba Research Europe
phone: +44 1223 436975

______________________________________________________________________
This email has been scanned by the MessageLabs Email Security System.
For more information please visit http://www.messagelabs.com/email 
______________________________________________________________________