[hts-users:02342] Re: Questions about brazilian demo

[Carlos Souza <carlosfss@xxxxxxxxx>]

Thanks a lot about your attention, Nicholas.
I'll take a look at this.

Regards,

--
Carlos Francisco Soares de Souza
Cel: (81)8635-1612 / (91)8831-5004
Mail: carlosfss@xxxxxxxxx
MSN: carlos@xxxxxxxxxxxxx
Skype: kafram



On Sun, Jan 3, 2010 at 8:54 AM, Nicholas Volk <nvolk@xxxxxxxxxx> wrote:
>> Hi,
>>
>> I've generated manually "full" and "mono" files, using an utt file, to
>> see how it works.
>> Yes, i've saw time information that you told me. When i ask you, i was
>> thinking that it may be part of string, and now i see that its a
>> output of dumpfeats.
>> The problem is in brazilian demo, cause it generate the "full" files
>> using a perl script, and it does not extract time information, only
>> generate the string.
>>
>> Now i've a big problem, i'll take a look in how can i use dumpfeats to
>> get this time information without the utts in this other (that i
>> believe is the standard) format, cause brazilian is so different.
>>
>> Thanks a lot, Nicholas,
>> Regards,
>>
>
> If you don't have a proper segmentation for your speech data you could use
> ehmm tool (part the Festvox package). Or calculate the wave length
> and use proportional lengths for phonemes. HTS internally will re-estimate
> the segment durations anyway so the quality of the original segmentation
> need not to be 100%.
>
> br,
>  Nicholas
>
>
>
>