[hts-users:00735] Re: Is it possible to fix corrupted .cmp files

["Nicholas Volk" <nvolk@xxxxxxxxxx>]

Hi,

For various reasons I've had periods of sound data that are all 0.
(I've manually silenced some clicks in the pause sections etc.)
This caused the +2126 error.

I was able to bypass this by creating a small program that adds random
very-low volume noise into the 0'd sections of the raw files
and the corrupted files work fine now.

But no matter what I do, I still get the same error +2126 when trying
to a 44100Hz voice, which is the original sample rate of my speech
database. (Same data fine when downsampled to 8000, 16000 and 22100.)
However this time there are now "nans" anywhere.

Has anyone on the list been able to compile a 44k voice?

br,
  Nicholas


> Hi,
>
> Zuko Fani wrote (2007/06/29 18:48):
>
>> I have found this 'nan' value in my .cmp files to have been responsible
>>
>> for the this error:
>> ERROR [+2126]  ViterbiAlign: No path found in 1'th segment
>>
>> Is it possible that it was caused by my wav to raw conversion
>> Or is it in extraction of mceps from raw files.
>> Or is it in the combining of mceps and If0 to .cmp.
>
> If all values in a frame are 0.0, extracted mel-cepstral coefficients
> become NaN.
> For example,
>
> % x2x +sf cmu_us_arctic_slt_a0001.raw | bcut -s 0 -e 399 | window -l 400
> -L 512 -w 0 -n 1 | mcep -a 0.42 -m 12 -l 512 | dmp +f
> 0       1.89489
> 1       0.822089
> 2       0.620753
> 3       0.66053
> 4       0.371214
> 5       0.548679
> 6       0.277446
> 7       0.43197
> 8       0.195604
> 9       0.194137
> 10      -0.0562655
> 11      0.0574899
> 12      -0.0505603
>
> % x2x +sf cmu_us_arctic_slt_a0001.raw | bcut -s 0 -e 399 | sopr -m 0.0 |
> window -l 400 -L 512 -w 0 -n 1 | mcep -a 0.42 -m 12 -l 512 | dmp +f
> 0       nan
> 1       nan
> 2       nan
> 3       nan
> 4       nan
> 5       nan
> 6       nan
> 7       nan
> 8       nan
> 9       nan
> 10      nan
> 11      nan
> 12      nan
>
> This problem sometimes occurs in silence.
> Please re-extract mel-cepstral coefficients from the raw file which caused
> this problem
> and check whether it includes NaN or not.
>
> 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://www.sp.nitech.ac.jp/~zen
> ------------------------------------------------
>