[hts-users:01465] Re: Trying new language on HTS

["Heiga ZEN (Byung Ha CHUN)" <zen@xxxxxxxxxxxxxxx>]

Hi,

Anocha Rugchatjaroen wrote (2008/06/25 16:19):

> We are trying Dzongkha language on HTS. What we have done is training a
> set of speech using HTS-demo_CMU_ARCTIC_SLT as a template without using
> DUMPFEATS of festival and setting up a training set as :
> 	- 53 phonemes ( consonants,  vowels) with 2 tones
> 	- 509 sentences (cover all diphone)
> 	- and automatic labeling using htk flat start.
> After we trained the script, we found that if we use full format of
> label file for generating speech - like
>
> 0       11900001 q_q-$+m=i/A:0
> 11900000        13300000 q_$-m+i=ph/A:0
> 13300000        15100000 $_m-i+ph=e/A:0
> 15100000        16100000 m_i-ph+e=l/A:0
> 16100000        17000000 i_ph-e+l=ph/A:0
> ...
>
> we can get a good synthesized speech. But if we use only
>
> q_q-$+m=i/A:0
> q_$-m+i=ph/A:0
> $_m-i+ph=e/A:0
> m_i-ph+e=l/A:0
> i_ph-e+l=ph/A:0
> ...
>
> we got very small speech - like a half of a syllable.
>
> Did we do anything wrong? Is our number of sentences too small?
> Could you please suggest us?

Did you use hts_engine or HMGenS?
Can you show us the command line arguments to generate speech parameters?

Regards,

Heiga ZEN (Byung Ha CHUN)

--
------------------------------------------------
Heiga ZEN     (in Japanese pronunciation)
Byung Ha CHUN (in Korean pronunciation)

Department of Computer Science and Engineering
Nagoya Institute of Technology
Gokiso-cho, Showa-ku, Nagoya 466-8555 Japan

http://www.sp.nitech.ac.jp/~zen
------------------------------------------------