[hts-users:02350] Re: Questions about brazilian demo

[Carlos Souza <carlosfss@xxxxxxxxx>]

Ranniery,

I've developed my own G2P, and it's very similar to this one you
recommended me, cause we based on the same main article.
I've made some adaptations, adding some rules, etc. I talked to author
of G2P of Ufpa, coincidentally i'm from Belém/PA too, despite of now
i'm living in Recife/PE.

I'll take a look at force-alignment with htk.. thanks for you help.

Let me ask you something. If this alignment is made using HTK, why
don't you have integrated this operation in brazilian demo of HTS?
like a perl script as you've done to produce the "data/labels/full"
files?
Is there a perl script to do this? or it's a more complex task?

Thank you again,
Regards,

--
Carlos F. S. Souza
carlosfss@xxxxxxxxx



On Mon, Jan 4, 2010 at 1:05 PM, Ranniery Maia
<ranniery.maia@xxxxxxxxxxxxxxxxx> wrote:
> Hi Carlos
>
> If you have your own G2P you may try running it on your database. Once
> you have your entire data labeled you can try forced alignment using HTK.
>
> This G2P for BP is free and its phone set is exactly the same as the one
> employed by the BP HTS-demo, except for the augmented phones /tS/ and /dZ/
> (the affricates):
>
> http://www.laps.ufpa.br/falabrasil/files/ConversorGrafema-fone1.3.rar
>
> I have never used Sphinx so I don't know if it works smooth or not.
>
> Hope it helps.
> Regards
>
> Ranniery
>
> ----------------------------
> Ranniery Maia
> Speech Technology Group
> Toshiba Research Europe LTD
> Tel: +44 1223 436974
>
>
>
> Carlos Souza wrote:
>>
>> Thanks for your reply, Ranniery,
>>
>> I'm happy to know that Brazilian demo is not a "dead one" as other guy
>> told. I'm trying to train HTS with another voice, but i'm having some
>> problems to create monophone label files.
>> Now i'm trying to create this with EHMM Labeller, but i still have no
>> sucess. Is there another way to create it? or using another tool, like
>> Sphinx?
>>
>> What's the better way to create this?
>>
>> Thanks about your attention,
>> Regards,
>>
>> --
>> Carlos F. S. Souza
>> carlosfss@xxxxxxxxx
>>
>>
>>
>> On Mon, Jan 4, 2010 at 6:48 AM, Ranniery Maia
>> <ranniery.maia@xxxxxxxxxxxxxxxxx> wrote:
>>>
>>> Hi
>>>
>>> There is nothing about "dead one" regarding the Brazilian demo. The
>>> intention is the best possible. The basic difference between the BP and
>>> English demos is that the
>>> latter uses Festival utterance files whereas the previous one has its own
>>> utterance file format. And that is because the text processor which
>>> generated the utterances
>>> files for BP is not free and there is nothing we can do about that.
>>> Therefore, if you intend to train HTS for a different BP voice you have
>>> to
>>> produce:
>>>
>>> 1) monophone label files with time boundaries and place them in
>>> data/labels/mono
>>>
>>> 2) use your own text processor to generate the utterance files and put
>>> them
>>> in data/utterinfo using the same format (which is not difficult once you
>>> have a G2P + a
>>> stress marker + a rough POS tagger).
>>>
>>> That should be just fine. But if you intend to wait, we have been
>>> negotiating with the HTS maintainer the release of a new version of the
>>> BP
>>> demo, this one based on
>>> Festival.
>>>
>>> Best regards
>>>
>>> Ranniery
>>>
>>> ----------------------------
>>> Ranniery Maia
>>> Speech Technology Group
>>> Toshiba Research Europe LTD
>>> Tel: +44 1223 436974
>>>
>>>
>>>
>>> Luis Uebel wrote:
>>>>
>>>> Oi Carlos,
>>>>
>>>> A pessoa responsável pelo  HTS para portugues nunca liberou o código ou
>>>> maiores explicações de como
>>>> fez o  sistema ou qual o T2P que utilizou.
>>>> É como os americanos chamam de dead one, ou seja, sem saída. Ele não
>>>> quer
>>>> ou não sabe como implementou.
>>>>
>>>>
>>>> Luís Felipe Uebel
>>>>
>>>>
>>>>
>>>>  > Date: Thu, 31 Dec 2009 10:05:50 -0300
>>>>  > From: carlosfss@xxxxxxxxx
>>>>  > Subject: [hts-users:02328] Questions about brazilian demo
>>>>  > To: hts-users@xxxxxxxxxxxxxxx
>>>>  >
>>>>  > Hi all,
>>>>  >
>>>>  > I've downloaded the brazilian demo of HTS and trainned it as well.
>>>>  > After that, when i was analysing and comparing it to english demo, i
>>>>  > saw that the utt files format are a kind of different.
>>>>  > I'd like to train HTS with another voice, that i recorded, and when i
>>>>  > tried to, i saw that files in folder "data/labels/mono" aren't
>>>>  > generated, and it's necessary to going on the process.
>>>>  > The brazilian demo still comes with this to their respective raws
>>>>  > files, and don't create or show how to get it.
>>>>  > English demo generate this ones using the utt files, that are in a
>>>>  > different format that is in brazilian's.
>>>>  >
>>>>  > Does anyone has faced this problem?
>>>>  >
>>>>  > Thanks a lot,
>>>>  > Regards,
>>>>  >
>>>>  > --
>>>>  > Carlos F. S. Souza
>>>>  > carlosfss@xxxxxxxxx
>>>>  >
>>>>
>>>> ------------------------------------------------------------------------
>>>> Add other email accounts to Hotmail in 3 easy steps. Find out how.
>>>> <http://clk.atdmt.com/UKM/go/186394593/direct/01/ >
>>>> ______________________________________________________________________
>>>> This email has been scanned by the MessageLabs Email Security System.
>>>> For more information please visit http://www.messagelabs.com/email
>>>> ______________________________________________________________________
>>>
>>> ______________________________________________________________________
>>> This email has been scanned by the MessageLabs Email Security System.
>>> For more information please visit http://www.messagelabs.com/email
>>> ______________________________________________________________________
>>>
>>>
>>
>>
>> ______________________________________________________________________
>> This email has been scanned by the MessageLabs Email Security System.
>> For more information please visit http://www.messagelabs.com/email
>> ______________________________________________________________________
>
>
> ______________________________________________________________________
> This email has been scanned by the MessageLabs Email Security System.
> For more information please visit http://www.messagelabs.com/email
> ______________________________________________________________________
>
>