Sir, I am working on building a tts using unit selection. I am replacing my auto labled .lab file with manually edited .lab files and then carrying out normal procedure to train the system the steps I used is given below after labelling the files.
$FESTVIVALDIR/bin/festival -b festvox/build_clunits.scm '(build_utts "etc/sp_unit_selection.data")' ./bin/make_pm_wave wav/*.wav > make_pm_wave.txt ./bin/make_mcep wav/*.wav > make_mcep.txt $FESTVIVALDIR/bin/festival -b --heap 10000000 festvox/build_clunits.scm '(build_clunits "etc/sp_unit_selection.data")' > build_clunits.txt ###PROSODY modeling### if data is too small then leave it. The system will work without it. ./bin/make_dur_model > make_dur_model.txt #for prosody ./bin/make_f0_model ./bin/make_f0_pm wav/*.wav #for prosody > make_f0_pm.txt Now after I completed the training, when I test with my trained data the sentences are not reading correctly, there were few pronunciation mistakes means it seems to me that they are using phone concatenation [similar to diphone concatenation]. I am totally new with this concept but as fas as I have read, in unit selection they search for sentence match else phrase , word at the end phones So my question is: Is there any reason for this error? I mean while labeling I am not indicating where is word break , phrase or sentence break Should I use this Break? Please it will be very kind of you if you could help to correct myself in this matter Thank You Shammur Absar Chowdhury
_______________________________________________ Festlang-talk mailing list Festlang-talk@lists.berlios.de https://lists.berlios.de/mailman/listinfo/festlang-talk
