[hts-users:01960] Re: How to configure HTS with STRAIGHT when using a different sampling rate?

[康世胤 <kangshiyin@xxxxxxxxx>]

The original date were sampled at 22050Hz. I used Sox to down-sample
the original data to 16kHz in the first experiment.  Recently I tried
to use the original data directly as shown in the second exp, and got
a very bad result.

On 4/29/09, Keiichiro Oura <uratec@xxxxxxxxxxxxxxx> wrote:
> Hi,
>
> What hertz were original speech signals were sampled at?
> What tool did you use for down-sampling to 22050Hz?
> How about 32kHz?
>
> Regards,
> Keiichiro Oura
>
> 康世胤 wrote:
>> hi
>>
>> I have build a voice by straight hts at 16Khz using the following param
>> , and the voice is good.
>>
>> SAMPFREQ    = 16000
>> FRAMESHIFT  = 80
>> FREQWARP    = 0.42
>> GAMMA       = 0
>> MGCORDER    = 39
>> MGCLSP      = 0
>> LNGAIN      = 1
>>
>>
>>
>> Now I am trying to build a voice at 22khz.
>> The following parameters are applied when configure, but  I got a very
>> bad voice.
>>
>> SAMPFREQ    = 22050
>> FRAMESHIFT  = 110
>> FREQWARP    = 0.45
>> GAMMA       = 0
>> MGCORDER    = 39
>> MGCLSP      = 0
>> LNGAIN      = 1
>>
>>
>> Should I change some other param to meet the new sampling rate ?
>> I have searched the mail list,
>> and found that the mgc order should also be changed when using a
>> differen sampling rate in hts without straight.
>>
>> So, should it be changed in straight hts?
>> Is  there a recommended value at 22khz?
>> Is this value OK?
>>         mgcorder = 39 / 16k * 22k = 53
>>
>>
>>
>>
>> --
>> 康世胤
>> Shiyin Kang
>>
>
>
>
>

-- 
Sent from Gmail for mobile | mobile.google.com

康世胤
Shiyin Kang

kangshiyin@xxxxxxxxx
Mobile: (+86)13810352340
Tel: (+8610)62775251

FIT 3-522
Lab of Human-Computer Speech Interaction,
Department of Computer Science & Technology,
Tsinghua University,Beijing, P.R.China, 100084