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

[hts-users:03484] Re: Question about HTS_lsp2mgc in hts_engine_API-1.06?

it should be alpha=0, not alpha!=0,the filter has delay-free loop? and what is the benefit of removing the  delay-free loop for the alpha!=0?

Keiichi Tokuda <tokuda@xxxxxxxxxxxx>编写:

When alpha != 0, the filter still has delay-free loops (even if gamma = -1).

2012/11/27 ArthurLeo <bin007.zhao@xxxxxxxxxxx>:
> I still confused. For the gamma=1,the mgc filter became lpc filter.but for general lpc filter,the c(m) is the lpc, it does not need covert to lpc to b(m), you can use general lpc filter directly. Is there delay-free loops in this general lpc filter such as AMR codec?
> 在 2012-11-27,1:14,"Keiichi Tokuda" <tokuda@xxxxxxxxxxxx> 写道:
>> The reason why c(m) is converted to b(m) is
>> to avoid delay-free loops in the digital filter structure.
>> Keiichi
>> 2012/11/26 ArthurLeo <bin007.zhao@xxxxxxxxxxx>:
>>> Thank you, I will double check the source code. I have the other question
>>> about c(m) and b(m).
>>> b(m) is the filter coefficients are used in the mglsa and mlsa filter(the
>>> attachment show the relationship between c(m) and b(m)). but why transform
>>> c(m) to b(m)? b(m) as the filter coefficients are much better than c(m)?
>>>> Date: Sun, 25 Nov 2012 16:06:06 +0000
>>>> From: heigazen@xxxxxxxxxx
>>>> Subject: [hts-users:03477] Re: Question about HTS_lsp2mgc in
>>>> hts_engine_API-1.06?
>>>> To: hts-users@xxxxxxxxxxxxxxx
>>>> Hi,
>>>> 2012/11/25 ArthurLeo <bin007.zhao@xxxxxxxxxxx>:
>>>>> it's first step is HTS_lsp2lpc,
>>>>> second is HTS_mgc2mgc(v, mgc, m, alpha, v->gamma, mgc, m, alpha,
>>>>> v->gamma);
>>>>> I can not understand why using the function mgc2mgc?the lsp2lpc function
>>>>> has transform LSP to lpc,why added the mgc2mgc?especially the parameters
>>>>> transformed are same?
>>>>> As far as I know,mgc2mgc transforms mel-generalized cepstral
>>>>> coefficients;
>>>>> mgc2mgc can transform frequency and generalized cepstral
>>>>> ([alpha1、gamma1]
>>>>> -》[alpha12、gamma2]),but now the transformed parameters are same? why?
>>>>> mgc2mgc function is necessary?
>>>> There are many forms of mel-generalized cepstral coefficients. For
>>>> example, mgcep supports 6 different forms of mel-generalized cepstral
>>>> coefficients as output (-o option, please refer to SPTK's reference
>>>> manual). However, mglsadf supports only a particular form of
>>>> mel-generalized cepstral coefficients (-o 0 in mgcep); mel-generalized
>>>> cepstral coefficients in the LPC form (-o 4 in mgcep if I remember
>>>> correctly) is not supported. Therefore, you need to convert
>>>> mel-generalized cepstral coefficients in the LPC form into those can
>>>> be used with mglsadf. HTS_mgc2mgc() performs this conversion.
>>>> Regards,
>>>> Heiga
>>>> --
>>>> Heiga ZEN (in Japanese)
>>>> Byung Ha CHUN (in Korean)
>>>> <heigazen@xxxxxxxxxx>