[hts-users:03073] Re: the order of CMLLR transforms being applied to models in the course of HTS engine generation
- Subject: [hts-users:03073] Re: the order of CMLLR transforms being applied to models in the course of HTS engine generation
- From: Hui LIANG <tshlmail-hts@xxxxxxxxx>
- Date: Wed, 19 Oct 2011 15:12:42 -0700 (PDT)
- Delivered-to: hts-users@xxxxxxxxxxxxxxx
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=yahoo.com; s=s1024; t=1319062362; bh=noNfaCtINGSWuayAtYb4tn8SeiifsYrUCdL55Lt/s3Q=; h=X-YMail-OSG:Received:X-Mailer:References:Message-ID:Date:From:Reply-To:Subject:To:In-Reply-To:MIME-Version:Content-Type:Content-Transfer-Encoding; b=C4Ov7WSL9lmSPDhb+VM647tkYANlvekqBiI+dqJYopYkYw8z2BohBTzMoRUXZv+7rLfuAPkbR4HNqru4rNP+b/2ClzCY9ul/nVMG6TdP87BtSXQN1lzTWpTTq8+bkXa8kigP5paSiGQHw8yhwHH8G6ObFqfYRkrSJJqM8211ZdI=
- Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=s1024; d=yahoo.com; h=X-YMail-OSG:Received:X-Mailer:References:Message-ID:Date:From:Reply-To:Subject:To:In-Reply-To:MIME-Version:Content-Type:Content-Transfer-Encoding; b=od6fQ1TWfym5Isv8xLUpta6htubWmB5Za87pNVJdMeg5hHhVmXViQ1+Dt4PEyqDDzmgnEsl3gO/OTsP/yRlmPHYIlHEhjbzckomH0Npg3jSfX6KoqimaRGQQjSa5Xsu/xTfq/i9nk5ax6n54RwKXOfiygL8ZthRcEsYAXwckxmU=;
Thanks a lot, 全さん. I tried the option -H, and found that -H and one more AX command resulted in the same HTS engine.
"HAdapt:ApplyCompXForm() first applies set_A then set_B."
This is exactly my concern. According to my understanding of the estimation/training of a cascade of CMLLR transforms, parent transforms (set_A) are applied to features and thus set_B should bridge the gap between speaker-independent models and adapted features. As a result, in the course of HTS engine generation, I feel that it is set_B that should be applied to the speaker-independent models first. In other words, set_B (or set_A) is "closer" to the model (or feature) side during training. set_B should be still "closer" to the model side during HTS engine generation.
Does swapping the two sets of CMLLR transforms make no/negligible difference? Or, my understanding is wrong?
Thanks,
Hui
----- Original Message -----
From: Heiga ZEN (Byung Ha CHUN) <heigazen@xxxxxxxxxx>
To: Hui LIANG <tshlmail-hts@xxxxxxxxx>
Cc: "hts-users@xxxxxxxxxxxxxxx" <hts-users@xxxxxxxxxxxxxxx>
Sent: Wednesday, 19 October 2011, 15:57
Subject: [hts-users:03072] Re: the order of CMLLR transforms being applied to models in the course of HTS engine generation
Hi,
2011/10/19 Hui LIANG <tshlmail-hts@xxxxxxxxx>:
> Suppose I have two sets of CMLLR transforms, set_A and set_B. set_A is parent transforms, meaning the following line
>
> <PARENTXFORM>~a "set_A"
>
> can be found in set_B.
>
> In the training phase, set_A is given and applied to features for the estimation of set_B (which bridges the gap between speaker-independent models and set_A-adapted features). But I am not sure whether HHEd firstly applies set_A or set_B to the speaker-independent models when I generate an HTS engine with the following commands:
>
> AX "set_A"
> AX "set_B"
> LT "(a decision tree)"
> CT "(output tree)"
> CM "(output engine file)"
I think
AX "set_B"
LT "(a decision tree)"
CT "(output tree)"
CM "(output engine file)"
is enough, as set_B has <PARENTXFORM>~a "set_A". HAdapt:ApplyCompXForm() first applies set_A then set_B. You may need to load set_A by -H option, though.
Regards,
Heiga
--
Heiga ZEN (in Japanese)
Byung Ha CHUN (in Korean)
<heigazen@xxxxxxxxxx>
- Follow-Ups
-
- [hts-users:03076] Re: the order of CMLLR transforms being applied to models in the course of HTS engine generation, Heiga ZEN (Byung Ha CHUN)
- References
-
- [hts-users:03071] the order of CMLLR transforms being applied to models in the course of HTS engine generation, Hui LIANG
- [hts-users:03072] Re: the order of CMLLR transforms being applied to models in the course of HTS engine generation, Heiga ZEN (Byung Ha CHUN)