I can't train thai language. about language-resources HOT 20 CLOSED

Seems there is still a issue with installation
because

/home/tontan/festival/build/festvox/src/ehmm/bin/FeatureExtraction: No such file or directory

FYI we have a docker image here https://hub.docker.com/r/langtech/base-festival

from language-resources.

Good job

One more suggestion, check the sample rate of your audio, if it is not 16000, resample them to be 16000 and train again.

from language-resources.

hmm seems that you might not have setup the festvox environment properly.

Try following this [1] again

1 - https://github.com/google/language-resources/tree/master/si/festvox#festvox-environment

from language-resources.

@pasindud It's not working.

Log https://gist.github.com/wannaphongcom/e08b4a371ef43412cbc39b5b549b8d39

from language-resources.

@pasindud Is it possible because of the installation problem?

Install log https://gist.github.com/wannaphongcom/3bde3d1d12512adfc9b81faaa1414eeb

from language-resources.

Yes it is a installation problem

from language-resources.

@pasindud http://tts.speech.cs.cmu.edu/awb/sptk/SPTK-3.6.tar.gz Not found.

from language-resources.

oh ok, Try
https://download.sourceforge.net/project/sp-tk/SPTK/SPTK-3.6/SPTK-3.6.tar.gz

from language-resources.

@pasindud It's not working. It's not train. https://gist.github.com/wannaphongcom/dba227ee8c835a0c8d6e87a0a9323ea7

from language-resources.

@pasindud I found a problem. Not found in lexicon but have words in lexicon. https://gist.github.com/wannaphongcom/7af873d5a489bce8cf3eb8494ec3eae9
Github : https://github.com/wannaphongcom/tts-thai/
audio : TSynC-1 https://www.nectec.or.th/corpus/index.php?league=sa

from language-resources.

Check whether the lexicon is sorted and uniq

Try the lexicon by

bash> festival
> (lex.set.compile.file "/path/to/lexicon.scm")
> (lex.lookup "<word to lookup for testing>")

More info http://www.festvox.org/docs/manual-2.4.0/festival_13.html

from language-resources.

@pasindud I try lexicon but It's show nil.

festival> (lex.set.compile.file "th/festvox/lexicon.scm")
"th/festvox/lexicon.scm"
festival> (lex.lookup "คน")
("คน" nil nil)
festival> (lex.lookup "คน")
("คน" nil nil)

lexicon.scm https://github.com/wannaphongcom/tts-thai/blob/master/festvox/lexicon.scm

from language-resources.

@pasindud Now, It works but I could not train.It has error.

Log https://gist.github.com/wannaphongcom/d633ad7af65cd7d18862c4cae8b2edf6

from language-resources.

What was the issue with the lexicon ?

Some letter is still not in the lexicon.

LEXICON: Word ด์ (plus features) not found in lexicon 

Can you post the festvox/___phoneset.scm

from language-resources.

@pasindud Do you mean phonology.json? https://github.com/wannaphongcom/tts-thai/blob/master/th/festvox/phonology.json

from language-resources.

@pasindud last train https://gist.github.com/wannaphongcom/c82d73ee9f1fac0a77eeb34cfb048247

from language-resources.

No, when you run the build script it creates a festvox folder, there is a file called festvox/___phoneset.scm in it

from language-resources.

@pasindud ok. https://gist.github.com/wannaphongcom/66d26af7bfc648ca4e9278675d91143e

from language-resources.

@pasindud Now, I trained it successfully but It has error when try synthesize one example sentence.

echo 'สวัสดี ทดสอบ ระบบ' |${FESTIVALDIR}/bin/text2wave \
> -eval festvox/goog_th_unison_cg.scm \
> -eval '(voice_goog_th_unison_cg)' example.wav
SIOD ERROR: could not open file example.wav

Train log and goog_th_unison_phoneset.scm : https://gist.github.com/wannaphongcom/07af6593577cdba7679e7dc50c0b50b3

from language-resources.

@pasindud Now, I found solution. I add -o to parameters.

echo 'สวัสดี ทดสอบ ระบบ' |${FESTIVALDIR}/bin/text2wave \
> -eval festvox/goog_th_unison_cg.scm \
> -eval '(voice_goog_th_unison_cg)' -o example.wav

audio : https://soundcloud.com/j5muzwuozouf/hello-tts

Thank you @pasindud

echo 'ขอบคุณ ครับ กูเกิล' |${FESTIVALDIR}/bin/text2wave \
> -eval festvox/goog_th_unison_cg.scm \
> -eval '(voice_goog_th_unison_cg)' -o thank_you_google.wav

audio : https://soundcloud.com/j5muzwuozouf/thank-you-google

Thank you Google!

from language-resources.