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

[hts-users:02464] Re: A question about gen directory (about adaptation)


·                                Dear all:

·                                 

·                                         I met the same problem, I do the adaptation use the demo script of hts-2.1 version with STRAIGHT,

·                                        

·                                         HERest find the transform matrix successfully (No error code appeared),

·                                 

·                                         but when I use HMGenS to generate the parameter sequence with transform matrix,

·                                 

·                                         from the synthesized speech I didn’t saw any adapted phenomenon.

                 

                 

 

                  Here is the command for generate the parameter sequence:

 

                  HMGenS -A -B -C syn.cnf -D -T 1 -S gen.scp -t 1500 100 5000 -h */*_*.lab -c 0

                   -H /mapmmf/re_clustered_all.mmf -N /dur/mapmmf/re_clustered_all.mmf

                   -M /gen/qst001/ver1/SI+dec_feat5/0

                   -a -J /cmp/xforms SI+dec_feat5 -H /cmp/regTrees/dec.base -H /ver1/cmp/regTrees/dec.tree

                   -b -Y /dur/xforms SI+dec_feat5 -N /dur/regTrees/dec.base -N /dur/regTrees/dec.tree /ver1/cmp/tiedlist /dur/tiedlist

 

 

 

 

 

                   In the almost same demo script without STRAIGHT I can successfully generate adapted speech,

                   But I input the transform matrix in Hts-engine not in the HMGenS.

 

·                                                                                                                

 

                  Best Regards, Daniel

 

·                                Subject: [hts-users:01863] A question about gen directory

·                                From: 최지성 <dir81@xxxxxxxxxxx>

·                                Date: Mon, 5 Jan 2009 02:06:28 +0000

·                                Delivered-to: hts-users@xxxxxxxxxxxxxxx

·                                Importance: Normal

Hi,

I tried to run the Speaker adaptation demo on my own data.

 

Processing directory

/home/jisung/HTS-demo-ADAPT/gen/qut001/ver1/SI/0  and

/home/jisung/HTS-demo-ADAPT/gen/qut001/ver1/SI+dec_feat3/0

 

synthesizing a  speech waveform fymn00001.wav from fymn00001.mgc, fymn00001.b0, fymn00001.dur, fymn00001.lf0, fymn00001.pit in /home/jisung/HTS-demo-ADAPT/gen/qut001/ver1/SI/0.

 

synthesizing an  adapted speech waveform fymn00001.wav from fymn00001.mgc, fymn00001.b0, fymn00001.dur, fymn00001.lf0, fymn00001.pit in /home/jisung/HTS-demo-ADAPT/gen/qut001/ver1/ SI+dec_feat3/0

 

I found that all of files have a same values for example fymn00001.mgc, fymn00001.b0, fymn00001.dur, fymn00001.lf0, fymn00001.pit in SI directory and SI+dec_feat3 directory (but speech files .wav are different)

 

I wonder why files are same values.   

 

SI                |           SI+dec_feat3

fymn00001.mgc   ß-same-à fymn00001.mgc

fymn00001.b0    ß-same-à fymn00001.b0

fymn00001.dur   ß-same-à fymn00001.dur          => Why???

fymn00001.lf0    ß-same-à fymn00001.lf0

.           ß-same-à      .

.           ß-same-à      .  

.           ß-same-à      .

=============================

fymn00001.wav    ß-differnet-à fymn00001.wav

 

 

Could you please help me a solution for this?

 

Thanks in advance for any hint

 

regards, Jisung

 


Follow-Ups
[hts-users:02465] HMM model by phoneme, Sébastien Le Maguer
References
[hts-users:02460] problem building regression-class trees and adaptation problem with xform mask, Alexandros Lazaridis