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

[hts-users:03440] Re: Problem with "forced alignment for no-silent GV" step of training.


Hi,

Some parameters such as FFT length, MGC order, etc. are set for 48kHz
in HTS demo scripts.
Please, change some parameters for 16kHz.

Regards,
Keiichiro Oura


2012/10/26 Martin Sulír <martin.sulir@xxxxxxxxx>:
> Hi all,
>
> I want to train my own voice with HTS STRAIGHT scripts, but I have a little
> problem in step called "forced alignment for no-silent GV".
>
> This step gives me an error:
>
> "WARNING [-9999]  HSMMAlign: No tokens survived to final node of network at
> beam [different numbers]".
>
> This step fail in most of .cmp/lab files and I dont obtain any data from it.
>
> When I use the same data (raw, utt, qst) in training with original HTS
> scripts (without Straight algoritm) I can train voice without this error.
> My speech data are sampled on 16kHz and I use configure arguments
> SAMPFREQ=16000 and FRAMESHIFT=80.
>
> How can I solve this problem?
>
> Thank you very much.
>
> Best Regards, M.S.
>
>
>

References
[hts-users:03438] Problem with "forced alignment for no-silent GV" step of training., Martin Sulír