[Subject Prev][Subject Next][Thread Prev][Thread Next][Date Index][Thread Index]

[hts-users:00127] Re: some questions about hts


From: eem1lz <eem1lz@xxxxxxxxxxxx>
Subject: [hts-users:00126] some questions about hts
Date: Wed, 27 Jul 2005 02:59:34 +0100
Message-ID: <FFF76EF083922F43AC90E794A73749F23F6D5A@xxxxxxxxxxxxxxxxxxxxxxxxxx>

> I am trying to use HTS to achieve speech recognition in one
> part of the project. but as a new user, I have no idea about
> HTS. so I put some quesitons here and hope someone help me.

The speech recognition part of the very low bit rate speech
coder can be constructed by using HTK.

> 1. how can I train a hidden markov model and get mean,
> variance, mix weight, transition matrix and model definition
> files by using HTS? (the model is 3-state with no skip)

Accordingly, the HTK Book, which can be downloaded from HTK web
site (http://htk.eng.cam.ac.uk/), should be referred to.

> 2. how can I get phoneme sequence and state durations from mel
> cepstrum by using phoneme recogizer , and then decode them
> back to mel cepstrum?

One of HTK tools, HVite, works as a phoneme recognizer with
appropriate settings for each language.

Heiga already posted a tool for generating a sequence of
mel-cepstrum from a state sequence to HTS-users ML.  See
[hts-users:00032] in the ML Archive:

  http://kt-lab.ics.nitech.ac.jp/hts-users/

Keiichi Tokuda
tokuda@xxxxxxxxxxxxxxxx
http://kt-lab.ics.nitech.ac.jp/~tokuda/

References
[hts-users:00126] some questions about hts, eem1lz