[hts-users:04254] Re: noise in generated voice

[Soheil Khorram <khorram.soheil@gmail.com>]

Dear Payman,

All parameters seem to extract correctly.

The mentioned noise is clearly the vocoder noise which can be removed by using STRAIGHT vocoder.

Moreover, The intonation of your sentence is clearly wrong. how many contextual factors have you used? Did you extract supra-segmental contextual factors (or at least POS tags)? 

You should use some supra-segmental factors to correct the intonation. 

 

Best Regards,

Soheil

On Sun, Apr 5, 2015 at 12:26 PM, payman shaykhmehdi <payman.shaykhmehdi@gmail.com> wrote:

Dear Soheil,

Thanks for your answer, 

A sample of synthesized sentence is attached, and before i use the STRAIGHT release, can you please explain the difference of these releases?

Thanks and Regards,

Payman Shaykhmehdi

On Sun, Apr 5, 2015 at 11:52 AM, Soheil Khorram <khorram.soheil@gmail.com> wrote:

Dear Payman,

The speech generated by normal HTS demo is a little buzzy.

Actually, the LPF is aimed to reduce the buzziness.

You could use the straight demo which is able to generate a synthesized speech with higher quality.

However, there are many possibilities that may cause this problem. if you send me a synthesized sentence of your voice, I can help you more.

Best Regards,

Soheil Khorram

On Sun, Apr 5, 2015 at 10:39 AM, payman shaykhmehdi <payman.shaykhmehdi@gmail.com> wrote:

hi all,

I am trying to generate persian voice using hts-demo. my database contain 938  wav files (48k). these files don't have any significant noise, but when i use htsvoice generated by hts-demo to synthesize a new sentence, the result contains a significant background noise. 

in training.pl script, i see that before making htsvoice, a low-pass filter applied to the result, can this filtering be the reason of that noise?

should i change some specific parameters?

or any other idea(s) for removing this kind of noise??

thanks and regards,

Payman Shaykhmehdi.