Did you do the train in one go?
Can we listen to any sample?
Try to train a voice with 100 files, it usually helps to see any errors. 100 files are usually enough to have some samples for each phone and it should sound something intelligible.
Regards2009/6/16 Tomasz Kuczmarski <faqster@xxxxxxxxx>They seem fine. I used the same limits for pitchmark extraction while building a clunits voice for Festival. At that time, PDA has successfully found pitch periods.2009/6/16 Tóth Bálint <toth.b@xxxxxxxxxxx>
Sorry for the basic question, but are the low and high limits of f0 extraction set fine?
Best Regards,
Bálint Tóth
Tomasz Kuczmarski írta:
The labelling is fine. I have hand checked the bigger part of the database. Moreover, I have built a unit selection synthesizer on this database and it works well.
I am just wondering whether eliminating any of the presented errors would help improve the quality.
--
Xavi.
Attachment:
Alice1.wav
Description: Wave audio
Attachment:
Alice2.wav
Description: Wave audio