[hts-users:03906] Re: Clustering all phones
- Subject: [hts-users:03906] Re: Clustering all phones
- From: Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>
- Date: Mon, 11 Nov 2013 11:28:36 +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:message-id:subject :from:to:cc:content-type; bh=FA+FFa2LZl6Xbn35RYHr+33LTFNuj74eVT/K/iLO0xw=; b=S7lKl1zO5uLdzm2T8U/pRlMVc+JgAd06ZMRssqqoq/tsTFWtiqaySwpVbmXRlSuHD6 SqDADJS+hV6c9GUR2AbrQvFGJjaJ8H6dk8CHFijidVaSq6bSIFzhziMUZDX+ooAV2Afz QjcPNqQSgYhMiPlH7/mqR9KqCAzDd1jkUVxAbx157kXGsMoraM7dVV13FLkZe6dWhBtc k+KNajBXiOGsq4iVrOrDx82Bz4kAiGcV/ZA1W+p0a7CZE3GeqqruBP1Q06KGIEpbFlJH GRagFt7cMzanK7eoOpNQPc9Vw6/tXsIT/D/ZVOElp1RmBEJLC2RKoMEHYzSF1BEcfQlO Foyg==
Hi,
As you said, in speech recognition, "current phoneme identity" context
is usually treated as a special context.
However, in HTS-demo scripts, "current phoneme identity" context is
regarded as one of the many context factors to be considered.
So, the similar phonemes can share some leafs.
Regards,
Keiichiro Oura
2013/11/11 Ibrahim Sobh <im_sobh@xxxxxxxxxxx>:
> Hi,
>
> Regarding clustering :
>
> Why we use:
> TB 0.00 mgc_s2_ {*.state[2].stream[1-1]} -->> state 2 from ALL phones
>
> and not use:
> TB 0.00 mgc_s2_ {phone.state[2].stream[1-1]} -->> state 2 from certain
> phone
>
> The reason could be because we have many context factors, however this will
> result in clustering states from totally different phones together! so how
> this really works?!
>
> Note: in ASR (HTK) we usually use "phone1.state[2], phone2.state[2] ....."
> for all phones.
>
> Regards
> Sobh
- Follow-Ups
-
- [hts-users:03910] Re: Clustering all phones, Ibrahim Sobh
- References
-
- [hts-users:03853] objective evaluation, Hea Young Park
- [hts-users:03857] Re: objective evaluation, Matt Shannon
- [hts-users:03858] Re: objective evaluation, Hea Young Park
- [hts-users:03871] HTS model and Speech Recognition, Ibrahim Sobh
- [hts-users:03872] Re: HTS model and Speech Recognition, Keiichiro Oura
- [hts-users:03893] Re: HTS model and Speech Recognition, Ibrahim Sobh
- [hts-users:03895] Re: HTS model and Speech Recognition, Keiichiro Oura
- [hts-users:03898] Change Speak Rate, Ibrahim Sobh
- [hts-users:03900] lf0, mgc, duration trees, Ibrahim Sobh
- [hts-users:03901] Clustering all phones, Ibrahim Sobh