[Subject Prev][Subject Next][Thread Prev][Thread Next][Date Index][Thread Index]

[hts-users:01202] Re: Error: LoadModelFiles: #duration pdf must be positive value.


Hi,

Thank you for your answer. I've checked the byte order of the pdf files, and it is big endian. The raw audio format is little-endian, signed, as in the CMU-ARCTIC-SLT.

Is it possible, that this error is caused by bad context-dependent labels or bad question file? Or what else can cause this problem?

Thank you for your answer in advance!

Best Regards,
Balint Toth

Heiga ZEN (Byung Ha CHUN) wrote:
Hi,

Tóth Bálint wrote (2008/03/01 2:02):

I'm trying to run Training.pl with my own speech database. Everything is ok until "Start synthesizing waveforms using hts_engine", I get the following error:

Synthesizing a speech waveform from /home/toth/work/HTS-demo/data/labels/gen/0001.lab using hts_engine...
Error: LoadModelFiles: #duration pdf must be positive value.
Error in /usr/local/bin//hts_engine -td /home/toth/work/HTS-demo/voices/qst/ver/tree-dur.inf -tf /home/toth/work/HTS-demo/voices/qst/ver/tree-lf0.inf -tm /home/toth/work/HTS-demo/voices/qst/ver/tree-mcp.inf -md /home/toth/work/HTS-demo/voices/qst/ver/dur.pdf -mf /home/toth/work/HTS-demo/voices/qst/ver/lf0.pdf -mm /home/toth/work/HTS-demo/voices/qst/ver/mcp.pdf -dm /home/toth/work/HTS-demo/voices/qst/ver/mcp.win1 -dm /home/toth/work/HTS-demo/voices/qst/ver/mcp.win2 -dm /home/toth/work/HTS-demo/voices/qst/ver/mcp.win3 -df /home/toth/work/HTS-demo/voices/qst/ver/lf0.win1 -df /home/toth/work/HTS-demo/voices/qst/ver/lf0.win2 -df /home/toth/work/HTS-demo/voices/qst/ver/lf0.win3 -s 16000 -p 80 -a 0.42 -b 0.4 -or /home/toth/work/HTS-demo/gen/qst/ver/hts_engine/0001.raw -ot /home/toth/work/HTS-demo/gen/qst/ver/hts_engine/0001.trace /home/toth/work/HTS-demo/data/labels/gen/0001.lab

With the example database (SLT) everything was working fine. What can be the problem?

Check endian of your *.pdf files.
In default, these files should be in big endian.

Regards,

Heiga ZEN (Byung Ha CHUN)



References
[hts-users:01190] How does HTK/HTS knows if the frame is voiced/unvoiced?, Antonio Bonafonte
[hts-users:01193] Re: How does HTK/HTS knows if the frame is voiced/unvoiced?, Heiga ZEN (Byung Ha CHUN)
[hts-users:01195] Re: How does HTK/HTS knows if the frame is voiced/unvoiced?, Antonio Bonafonte
[hts-users:01196] Error: LoadModelFiles: #duration pdf must be positive value., Tóth Bálint
[hts-users:01197] Re: Error: LoadModelFiles: #duration pdf must be positive value., Heiga ZEN (Byung Ha CHUN)