[hts-users:03366] Re: Regarding GV for spectrum
- Subject: [hts-users:03366] Re: Regarding GV for spectrum
- From: Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>
- Date: Mon, 25 Jun 2012 19:13:27 +0900
- Cc: uratec <uratec@xxxxxxxxxxxx>
- Delivered-to: hts-users@xxxxxxxxxxxxxxx
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=mime-version:sender:in-reply-to:references:date :x-google-sender-auth:message-id:subject:from:to:cc:content-type :content-transfer-encoding; bh=26M8mNttbq9EvAcF6EQrEnzAgIUrhlr0TUmZdzJ8u1w=; b=rqFKQyh6eWD8nE5qvrYe+h2GZrJNvNFXNLcjLvGXBHuhdjERB3mGQ6yJkULYFKaXAM m1yE43met0IxI/u60leVvZH3GpmCyQ7sky+l4+DSwjvhR1SHKxDXcikel0hHfh5r2EdB ifHUaciQVkiivncdshU2BfFenaR5TT8Wr09i0aznJOXQ0MzICT97auAOcJ9xxImWKw92 uawLIudzXLVv3grnR7Tqu16d3Rh+eLskydvaaQ9G88sa3PkxZQrlcZ64SMU/fwReT8YM 9B/h+9xeWE9owGSd2IJ3CrV46yyBB5iYg0SPCCBvE8AKuYVGT8X6+53zUwvnPtjjcOtd FqHQ==
Hi,
You can train speaker-dependent GV by using `Speaker dependent
training demo' script.
http://hts.sp.nitech.ac.jp/?Download
Regards,
Keiichiro Oura
2012/6/25 Kwan Lisa <lisakwan1102@xxxxxxxxx>:
> Hi,
>
> I thought I was using the GV model of the target speaker. But I am not
> sure what kind of GV model I was using.
> I found that when I use HMGenS to generate the voice for the speaker
> independent(SI) model, my GV setting is:
> GVHMMLIST /home/rex/TTS/grad/TTS2/AST_realign_20speakers_5tone/gv/tiedlist
> GVMODELMMF
> /home/rex/TTS/grad/TTS2/AST_realign_20speakers_5tone/gv/clustered_all.mmf
>
> When I use HMGenS to generate the voice for the speaker dependent(SD)
> model, my GV setting is still:
> GVHMMLIST /home/rex/TTS/grad/TTS2/AST_realign_20speakers_5tone/gv/tiedlist
> GVMODELMMF
> /home/rex/TTS/grad/TTS2/AST_realign_20speakers_5tone/gv/clustered_all.mmf
>
> According to my logs, it seemed that I didn't train GV model for the
> target speaker. So I think the GV model I was using was the one of SI
> model.
> How should I train a speaker dependent GV model? I thought the GV
> model used in the demo script was re-trained using adaptation data.
>
> 2012/6/25 Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>:
>> Hi,
>>
>> I guess you use speaker-independent GV model.
>> If there are many adaptation utterances, could you try to calculate
>> target speaker-dependent GV?
>>
>> Regards,
>> Keiichiro Oura
>>
>>
>> 2012/6/22 Kwan Lisa <lisakwan1102@xxxxxxxxx>:
>>> Hi,
>>>
>>> I have question about the weight of GV for spectrum. When synthesis a
>>> voice with hts-engine, there is a option controlling the weight of GV
>>> for spectrum. I am using a kind of average voice model to adapt to the
>>> target speaker, and I found when I tune the weight higher, the voice
>>> is more similar to the target speaker. But there was a trade-off. The
>>> higher the weight, the more explosive sound in the voice. I'm not sure
>>> why it happened. How do I diminish the explosive noisy sound in the
>>> voice?
>>>
>>> --
>>> Lisa Kwan
>>> lisakwan1102(at)gmail.com
>>> Advanced Speech Technology Lab, ASTL
>>>
>>
>
>
>
> --
> Lisa Kwan
> lisakwan1102(at)gmail.com
> Advanced Speech Technology Lab, ASTL
>
- References
-
- [hts-users:03362] Regarding GV for spectrum, Kwan Lisa
- [hts-users:03363] Re: Regarding GV for spectrum, Keiichiro Oura
- [hts-users:03365] Re: Regarding GV for spectrum, Kwan Lisa