[hts-users:04389] Re: Building Chinese voices using HTS
- Subject: [hts-users:04389] Re: Building Chinese voices using HTS
- From: HE Qiang <qhe1228@xxxxxxxxx>
- Date: Sat, 19 Mar 2016 14:01:49 +0800
- Delivered-to: hts-users@xxxxxxxxxxxxxxx
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=mime-version:in-reply-to:references:date:message-id:subject:from:to; bh=fNPQpiNClrRBUz0RZaeP2iLsqtY1vLXHLwMGyfUcK7k=; b=phR3AYltE6LEUYmjV1zqBUeVPNPFyM8rhDV8uSNyA5fv9I7rP+w8/O3ygt9+x+yQo6 4XtWjQ7kfhQENklQ+ORDayygfc3E7P4/2g2zTpP/p7TcIcI54ZcaZHyTrvNCxSWrImiF Yj5ClBKw38jsT4Rla/MDPzstEzdS3E+gq6gmJ/40KTjrGv1Hk4cGjC5Fdy9QK2VGXrTg Ldc5qeXhtom9Yd+vEAIBbBZYQ9E1yB+zFgHe+O9eOHer/U6ci27uOmSUgM0j7VVUtgxv bJx7/ZbGKPAysdgTc+kPi8ZWOKaKKZnE2V3ljx7Ogqz3z+bNwlgvjZangr+xI8ao1wSI vSpA==
Dear Dr. Chen:
I set up a mandarin HTS experiment system a few years ago, and finally
build a win32 console program to convert text pinyin sentence to
mandarin wave file. The program is as small as 436K bytes.
I think the first and most important step to build a mandarin HTS
system is to prepare a hed file contains a lot of questions, like:
QS "C_sil" {*-sil+*}
QS "C_sp" {*-sp+*}
QS "L_w" {w-*}
QS "L_yv" {yv-*}
QS "R_b" {*+b/A*}
QS "R_d" {*+d/A*}
QS "R_yv" {*+yv/A*}
QS "C_Sheng_Sai" {*-b+*, *-d+*, *-g+*, *-p+*, *-t+*, *-k+*}
QS "C_Sheng_SongQiSai" {*-b+*, *-d+*, *-g+*}
to make such a hed file, you'd better write a script to automatically
generate it, like mine:
q_sheng_all={...
'Sheng_All', {'b', 'd', 'g', 'p', 't', 'k', 'z' , 'zh',
'j', 'c', 'ch', 'q', 'f', 's', 'sh', 'xx', 'h', 'r', 'm', 'n', 'l',
'oa', 'oo', 'oe', 'yi', 'w', 'yv'};
};
q_sheng={...
'Sheng_Sai', {'b', 'd', 'g', 'p', 't', 'k'};
'Sheng_SongQiSai', {'b', 'd', 'g'};
...
%current is sil,sp
print_triphone('C', '*-%s+*', 'sil', 'sil')
print_triphone('C', '*-%s+*', 'sp', 'sp')
print_triphone('C', '*-%s+*', 'sil_sp', {'sil','sp'})
That are a lot of other trivial work.
Best wishs,
Qiang HE
2016-03-16 16:28 GMT+08:00, DONGPENG CHEN <dpchen@xxxxxxxxxx>:
> Hi,
>
> I worked on ASR for many years, but recently we need to build Chinese
> voices.
>
> I have successfully run the HTS demo for English. I found that HTS actually
> calls the festvox to convert *.txt files to *.utt files. If I am right,
> festvox makes use of cmu English dictionary to do this.
>
> My question is, if I have Chinese waves, labels, and dictionary, how can I
> make festvox convert my *.txt to *.utt ? Or is there tutorial on how to set
> up new languages using HTS?
>
> Another concern is, do I need to add questions on tones for the Chinese
> decision tree since it is a toning language?
>
> I know these are old questions, but I can't find answer from the history
> threads (forgive my poor experience on TTS). Hope you can help me.
>
> Thanks a lot.
>
> Best wishes,
> D. Chen
>
- References
-
- [hts-users:04387] Building Chinese voices using HTS, DONGPENG CHEN