Closed xyx361100238 closed 2 years ago
@ZackHodari
Your issue seems to be related to HTS, forced alignment, or your data, not merlin.
I was use merlin to force alignment : ./scripts/submit.sh ${MerlinDir}/src/run_merlin.py conf/test_dur_synth_mandarin_voice.conf Actually,there is silence section in train wav file,so the format like:
0 400000 xx^xx-sil+j=i4@xx@/A:xx-xx^xx@/B:xx+xx@xx^xx^xx+xx#xx-xx-/C:……
400000 900000 xx^sil-j+i4=x@i@/A:xx-4^4@/B:0+19@1^2^1+20#1-20-/C:xx_v……
900000 1600000 sil^j-i4+x=v4@i@/A:xx-4^4@/B:0+19@1^2^1+20#1-20-/……
Dear ALL: I was use MTTS to deal Mandarin TTS frontend (text2label with no dur):
then use merlin to alignment:
When I use this label files to train in HTS,tips:
it means no have enough lab for phone 'sil',that's because dur_label is 0 ~ 0 in line one。 how to slove this problem,thinks
(it works well in merlin system)