[hts-users:04341] Re: about using HMGenS to synthesized speech with desired duration

[Alexis Moinet <alexis.moinet@xxxxxxxxxxx>]

PS: according to the link given previously by Bajibabu Bollepalli:

Change the function call to HMGenS in the Training.pl script to use state-level alignment labels by adding –s parameter and remove the duration model:

So definitely "-s" (and "-m" for models) is what you're looking for.

On 03/12/15 15:51, Alexis Moinet wrote:

If you run HMGenS with no parameters in command-line, you should get a list of all available options for it.

Two of them seem of interest: "-m" and "-s" seem to enable forced alignment (ala -vp) either using models or state duration as given in labels (I've never tested these options though, so I might have misinterpreted the help message)

Hope it helps,

Alexis

On 03/12/15 15:42, Tuan Dinh wrote:

Thanks you very much. I very appreciate.

But I expect something like "-vp" option of hts_engine.

Isn't any similar option with HMMGenS to use pre-specified duration of phonemes.

Best regards.

-Tuan

2015-12-03 22:29 GMT+09:00 Bajibabu Bollepalli <bajibabu7@xxxxxxxxx>:

Hello,

This post may help you http://www.nguyenquyhy.com/2013/12/how-to-configure-hts-for-in-training-synthesis-with-state-level-alignment-labels/

/Baji

On 3 December 2015 at 15:03, Tuan Dinh <tuan.dinh@xxxxxxxxx> wrote:

Hi all,

I have label files with duration information. How can I make HMGenS use the duration information in synthesizing speech? Thank  you very much.

Best regards,

-Tuan