I constructed a voice with a database of 9h of speech, and succeded. However, I changed the question list (I added some new questions) and now in the step "start making unseen models (stc)" I get the error:
HModel: checking HMM n^e-n+e=l~2_1/A:0_1_4/B:0-0-2~1-1&8-16#0-0$2-3!0-0;1-6|e/C:0+0+2/D:NC_3/E:SP+1~4+7&2+3#1+2/F:DA_1/G:0_0/H:23=10~1=1|F/I:0_0/J:11+24-1
HModel: checking state 2, si=0x7f9a3d8e2380
HModel: checking stream 2, se=0x7f9a4118cb38
HModel: checking mix 1
HModel: checking mix 2
HModel: checking stream 3, se=0x7f9a4118cde8
HModel: checking mix 1
ERROR [+5220] CovDet: [-2198934781952.000000 ...] not invertible
FATAL ERROR - Terminating program /veu4/usuaris20/tts/tools/hts/bin/HHEd
It might be because of a lack of training data in estimating the model parameters, but how can I check the amount of data use to train each model? Would someone give me a hint on this?
Regards.
--
Jordi Adell
http://www.jordiadell.org