[hts-users:02575] Re: Error in synthesizing waveforms using hts_engine step

[那兴宇 <nxy-yzqs@xxxxxxx>]

Hi, Igor

 

1mix synthesis used the HMGens tool and SPTK (or plus STRAIGHT). But hts_engine is a isolated module. They use different types of model.

In hts_engine, the models in dir "voices" is used. Binary file "lf0.pdf" stores the means, variance and msd flag of lf0.

This error happens when the synthesis engine load that model file, which means it read of a msd flag out of 0.0 to 1.0.
Perhaps something happened during the conversion of the model set format. You can check that step.

 

Una abrazo,

Na Xingyu

Beijing Institue of Technology

At 2010-08-03 01:51:45，Igor <igorcouto@xxxxxxxxx> wrote:
> Hi, 
>
> I am facing a problem in HMM training. When the script runs the comand:
>
> /home/04080004701/tools/hts_engine/bin/hts_engine -td /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/tree-dur.inf -tf /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/tree-lf0.inf -tm /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/tree-mgc.inf -md /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/dur.pdf -mf /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/lf0.pdf -mm /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/mgc.pdf -dm /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/mgc.win1 -dm /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/mgc.win2 -dm /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/mgc.win3 -df /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/lf0.win1 -df /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/lf0.win2 -df /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/lf0.win3 -s 16000 -p 80 -a 0.42 -g 0  -b 0.4 -cm /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/gv-mgc.pdf -cf /home/04080004701/voices/brazilian_male_voice/hts/voices/qst001/ver1/gv-lf0.pdf -b 0.0  -or /home/04080004701/voices/brazilian_male_voice/hts/gen/qst001/ver1/hts_engine/gen_portuguese_m001_024.raw -ot /home/04080004701/voices/brazilian_male_voice/hts/gen/qst001/ver1/hts_engine/gen_portuguese_m001_024.trace /home/04080004701/voices/brazilian_male_voice/hts/data/labels/gen/gen_portuguese_m001_024.lab
>
> The following error is returned:
>
> Error: HTS_Model_load_pdf: MSD weight should be within 0.0 to 1.0.
>
> It it strange because, because the previous steps, like synthesizing waveforms (1mix), are processed normally.
>
> Someone has a solution or a hint?
>
> Thanks.
>
> -- 
> Eng. Igor Costa do Couto
> Engenharia da Computação - UFPa
> LaPS - UFPa
> igorcouto@xxxxxxxxx
> Skype: iccouto

---

网易邮箱，没有垃圾邮件的邮箱。