[hts-users:00131] Re: about training files
- Subject: [hts-users:00131] Re: about training files
- From: "Heiga ZEN (Byung Ha CHUN)" <zen@xxxxxxxxxxxxxxxx>
- Date: Fri, 19 Aug 2005 13:06:51 +0900
- Organization: Nagoya Institute of Technology, Japan
- User-agent: Mozilla Thunderbird 1.0.2 (Windows/20050317)
Hi,
eem1lz wrote:
I am trying to build phoneme HMMs for speech recognition.
The training file I am using is the mcep file (24 order) generated by the SPTK.
But the mcep is not the standard HTK file so that it can not be usd in HTK directly.
So how can I generate the dynamic feature file (75 dimensional file with delta and double delta) from the mcep file and
then convert it to the parametered file with the header which can be read by HTK?
The question may seems simple, but I really need your help,
> please give tips or a sample, since HTK book does not contain much
information about the conversion between the user-definied files)
At script directory in HTS-demo_CMU-ARCTIC-AWB.tar.gz which is
distributed in the HTS web site, you can find mkdata.in (or mkdata.pl).
This script is designed to prepare training data from waveforms.
In this script, you can find
perl $scpdir/delta.pl $mcepdim $mcepdir/${base}.mcep @mcepwin >
$cmpdir/tmp.mcep
This script reads regression window coefficients, computes dynamic
features, and outputs results.
So please run this script with your input/output mcep filenames as follows:
perl scripts/delta.pl 25 aaaaaa.mcep win/mcep_dyn.win win/mcep_acc.win >
aaaaa.mcep_dyn_acc
You can also find following lines in the mkdata.in (or mkdata.pl).
@STAT = stat "$cmpdir/tmp.cmp";
$size = $STAT[7]/$byte;
system "echo $size ".($frameshift * 10000000)." | $SPTKdir/x2x +al >
$cmpdir/tmp.head";
system "echo $byte 9 | $SPTKdir/x2x +as >> $cmpdir/tmp.head"
system "cat $cmpdir/tmp.head $cmpdir/tmp.cmp > $cmpdir/${base}.cmp";
In these lines, HTK header is prepared and concatenated to the training
data.
By using them, you can attach HTK header to your own data.
Regards,
Heiga Zen (Byung Ha CHUN)
--
------------------------------------------------
Heiga ZEN (in Japanese pronunciation)
Byung-Ha CHUN (in Korean pronunciation)
Department of Computer Science and Engineering
Graduate School of Engineering
Nagoya Institute of Technology
Japan
e-mail: zen@xxxxxxxxxxxxxxxx
web: http://kt-lab.ics.nitech.ac.jp/~zen
------------------------------------------------
- References
-
- about training files, eem1lz