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

[hts-users:02994] Re: Silent waveforms in synthesis with hts_engine


Hi, Chaomin,
       I want to know  how did you add the ap stream to "hts_engine API".  Did you modify the source code?
       In addition, could you tell me how to add the energy to the MCEP. 

Regards,

hlwang

At 2011-08-08 13:12:53,"王朝民" <wangchaomin@xxxxxxx> wrote:
Hi,
    i added the ap stream to "hts_engine API", and it's run ok. And  yours silent waveforms are not really silent when you tune the scale of the amplitute ,you can see the waveform exist. So as to Oura, the main difference between mel-cepstrum from SPTK and STRAIGHT is the scale, because of the Matalb process of waveform. So you can add the anergy to the mcep yourself when processing.
   hope my answer can help you.
 
                                                                                              Chaomin Wang  ( 王朝民)


At 2011-08-08 11:13:42,"Keiichiro Oura" <uratec@xxxxxxxxxxxxxxx> wrote: >Hi, > >The main difference between mel-cepstrum from SPTK and STRAIGHT >(Matlab) is the scale. >In STRAIGHT, a type of normalization is implemented. >But, I don't remember the details. > >Anyway, "hts_engine API" doesn't support aperiodic measure. >Even if you solve the above problem, synthesized speech is not STRAIGHT voice. > >Regards, >Keiichiro Oura > > >2011/8/8 huanliangwang <huanliangwang@xxxxxxx>: >> Hi, Oura: >>      "hts_engine API'' supports MCEP and f0 feature to synthesize speech. >> But why dose it not support MCEP and f0 feature (excluding bap feature ) >> generated from STRAIGHT to synthesize speech? What is the main difference >> between these features? >> >> Thanks >> >> >> hlwang >> >> >> >> >> >> >> >> >> >> At 2011-08-08 09:22:25,"Keiichiro Oura" <uratec@xxxxxxxxxxxxxxx> wrote: >>>Hi, >>> >>>``hts_engine API'' doesn't support STRAIGHT. >>>Please, use HMGenS and Matlab. >>> >>>Regards, >>>Keiichiro Oura >>> >>> >>>2011/8/8 huanliangwang <huanliangwang@xxxxxxx>: >>>> Hi, >>>>        I also met with similar problem when using hts_engine to synthesize >>>> wave in HTS-demo_CMU-ARCTIC-SLT_STRAIGHT. >>>> Dose anyone know how to solve the problem or the reason it occurs? >>>> >>>> Regards, >>>> >>>> hlwang >>>> >>>> At 2011-08-08 01:49:09,"Kathrin Haag" <kathrin_haag@xxxxxxxxxxx> wrote: >>>> >>>> Hello, >>>> >>>> when I synthesise waveforms with hts_engine, I don't receive any errors, but >>>> get waveforms without any signal as the output. >>>> I used hts_engine for a different voice and got the correct output! I also >>>> have all models I usually need for synthesis! >>>> >>>> I compared the wavefile sizes of the silent waveforms to the wavefile sizes >>>> of my correctly synthesised waveforms from a different voice, and they all >>>> have approximately the same size. >>>> >>>> Since I did not receive any substantial errors during training, I assume >>>> that the error occurred during feature extraction at synthesis time. I >>>> analysed the synthesised waveforms, and they don't have any feature values. >>>> >>>> I'd appreciate any suggestions on how to solve this problem. >>>> >>>> Thanks, >>>> >>>> Kathrin Haag >>>> >>>> >>>> >>>> >>>> >>>> >>> >> >> >> >





References
[hts-users:02992] Re: Silent waveforms in synthesis with hts_engine, 王朝民
[hts-users:02990] Re: Silent waveforms in synthesis with hts_engine, Keiichiro Oura
[hts-users:02986] Silent waveforms in synthesis with hts_engine, Kathrin Haag
[hts-users:02987] Re: Silent waveforms in synthesis with hts_engine, huanliangwang
[hts-users:02988] Re: Silent waveforms in synthesis with hts_engine, Keiichiro Oura
[hts-users:02989] Re: Silent waveforms in synthesis with hts_engine, huanliangwang