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

[hts-users:02843] Re: Problem of using lspdf to synthesize


Model training is not involved in my experiment yet.

I'm just learning to use the SPTK tools 'lpc', 'lpc2lsp', 'poledf', 'lspdf', just like the examples in Chapter 4 of "SPTKexamples-3.4.1.pdf", which has been attached.

Thus, the 'lsp.data' was obtained using 'lpc2lsp' derectly, and the 'lpc.data' obtained by analysing the basicly processed test signal.

'Original.wav'---->basic processing(frame, window)---->lpc---->lpc2lsp---->lspdf----->'lspsyn.wav'
          |                                                                                                                       |
          |------->pitch----------->excite -------------------------------------------------------------------------->|



在 2011年5月16日 下午5:41,那兴宇 <nxy-yzqs@xxxxxxx>写道:
No. GV option is for the parameter generation stage, i.e., the step generated your lsp.data.
LSP may cause some unsteady frames which will form large formant peaks.

At 2011-05-16 14:39:32,"Shijun Qian" <sjqian@xxxxxxxxx> wrote:
Sorry! I reload the lspsyn_m24.wav!

It happens at the voice frames. And there seems no GV option in the lspdf.


在 2011年5月16日 上午11:48,那兴宇 <nxy-yzqs@xxxxxxx>写道:
You misloaded the lspsyn voice.
Does it happen at the voice frames or the silence frames? Did you turn on GV option when synthesizing?

At 2011-05-13 14:00:27,"Shijun Qian" <sjqian@xxxxxxxxx> wrote:
Yes, I already tried the 'lspcheck', and there's no problem with the 'lsp.data'.

Besides, I also tried to use the 'lsp2lpc' to transfer 'lsp.data' back to 'lpc.data', and use 'poledf' to resynthesize, and there's no problem with the resynthesized data.

在 2011年5月13日 下午1:38,Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>写道:
Hi,

Could you try 'lspcheck -r' in SPTK?

Regards,
Keiichiro Oura


2011/5/13 Shijun Qian <sjqian@xxxxxxxxx>:
> There was no frequency warping factor problem when I encountered this lspdf
> trouble.
>
> I got the LSP coefficients from the LPC coefficients:
>
> -------------
> lpc -l 512 -m 24 windowed.data > lpc.data
>
> lpc2lsp -m 24 -n 256 lpc.data > lsp.data
>
> poledf -m 24 -p 80 lpc.data excite.data > lpcsyn.data
>
> lspdf -m 24 -p 80 lsp.data excite.data > lspsyn.data
> -------------
>
> It turns out that the lpcsyn.data is ok, while the lspsyn.data has the
> problem I mentioned -- some synthesized points' value turn out to be
> extremely large.
>
> However, the problem disappers when m<20.
>
> I attached the three wave files with the order m=24: original.wav,
> lpcsyn_m24.wav, lspsyn_m24.wav.
>
> Please help me to solve the problem.
>
>
> 在 2011年5月13日 上午9:42,那兴宇 <nxy-yzqs@xxxxxxx>写道:
>>
>> When using lspdf, the frequency warping factor should be zero.
>>
>> At 2011-05-11 23:31:57,"Shijun Qian" <sjqian@xxxxxxxxx> wrote:
>>
>> Hello!
>> I encountered a problem when using lspdf for synthesis:
>> When the lsp order is relatively large, e.g. m>20, some synthesized
>> points' value turn out to be extremely large, which makes the synthesized
>> speech sound annoying. And larger the lsp order, more the abnormal
>> synthesized points.
>>
>> --
>> Sincerely,
>> Shijun Qian (钱诗君)
>> Wireless Networks Laboratory
>> Nangong Rm210, SEU Research Institute
>> No.399, Linquan Street, Suzhou, 215123, P. R. China
>>
>>
>
>
>
> --
> Sincerely,
> Shijun Qian (钱诗君)
> Wireless Networks Laboratory
> Nangong Rm210, SEU Research Institute
> No.399, Linquan Street, Suzhou, 215123, P. R. China
> Mobile: +86 15606201626
>




--
Sincerely,

Shijun Qian (钱诗君)
Wireless Networks Laboratory
Nangong Rm210, SEU Research Institute
No.399, Linquan Street, Suzhou, 215123, P. R. China
Mobile: +86 15606201626






--
Sincerely,

Shijun Qian (钱诗君)
Wireless Networks Laboratory
Nangong Rm210, SEU Research Institute
No.399, Linquan Street, Suzhou, 215123, P. R. China
Mobile: +86 15606201626






--
Sincerely,

Shijun Qian (钱诗君)
Wireless Networks Laboratory
Nangong Rm210, SEU Research Institute
No.399, Linquan Street, Suzhou, 215123, P. R. China
Mobile: +86 15606201626


References
[hts-users:02828] Problem of using lspdf to synthesize, Shijun Qian
[hts-users:02831] Re: Problem of using lspdf to synthesize, 那兴宇
[hts-users:02832] Re: Problem of using lspdf to synthesize, Shijun Qian
[hts-users:02834] Re: Problem of using lspdf to synthesize, Keiichiro Oura
[hts-users:02835] Re: Problem of using lspdf to synthesize, Shijun Qian
[hts-users:02836] Re: Problem of using lspdf to synthesize, 那兴宇
[hts-users:02839] Re: Problem of using lspdf to synthesize, Shijun Qian
[hts-users:02840] Re: Problem of using lspdf to synthesize, 那兴宇