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

[hts-users:00285] Re: HTS and STRAIGHT


Thanks a lot. If I get things right, you have written some additional
code to STRAIGHT to achieve what you describe below. Is it possible for
me to get this code? It would be of great help. :)

sincerely,

Lene


Sitat "Heiga ZEN (Byung Ha CHUN)" <zen@xxxxxxxxxxxxxxxx>:

> Hi,
>
> lenemo@xxxxxxxxxxxx wrote:
>
> > In the first instance I'm interested in using STRAIGHT to extract
> the
> > speech parameters and let this feature vector be used to train
> HMM's
> > with HTS. Do you have any clue on how this can be done? What
> changes do
> > I have to do in the scripts?
>
> Usually, STRAIGHT extracts F0, F0-adaptively-smoothed amplitude
> spectrum (512 point), aperiodicity ratio values in
> frequency domain (512 point).
> In the Eurospeech 2005 paper, we have extracted mel-cepstral
> coefficients from this F0-adaptively-smoothed amplitude
> spectrum then used them instead of normal mel-cepstral coefficients.
> And we have averaged aperiodicity ratio values over 5 frequency
> sub-bands.
>
> After appending delta and delta-delta to F0 (1 dimensions),
> STRAIGHT-mel-cepstrum (40 dimensions), and band-averaged
> aperiodicity ratio values (5 dimensions), they were composed and
> final feature vectors (138 dimension) were generated.
>
> By the way, can you send any questions about HTS to the hts-users ML
> This kind of information could be valuable for all people who are
> interested in HTS.
>
> Best 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://kt-lab.ics.nitech.ac.jp/~zen
> ------------------------------------------------
>