[mary-users] HMMVoiceMakeData - PhoneFeatureLabelAligner - HMMVoiceMakeVoice

[Ingmar Steiner]

Dear Florent,

On 04.08.2011 01:06, fxavier at ircam.fr wrote:
> Dear all,
>
> I'm very happy to announce you that we finally manage to have the french
> voice!

Congratulations on getting the voice built! =)

>
> - PhoneFeatureLabelAligner: Alignement problem was because some pauses
> missed at the end of the lab file in /lab directory. Solution, don't run
> the PhoneUnitLabelComputer, an simply copy/paste the /lab/*.lab into
> /phonelab/*.lab

I'm pretty sure that this would have solved itself if all the necessary 
components had been run in the right order... For instance, the pfeats 
files would not have contained extra trailing _ units if they had been 
computed from lab files without those extra segments. It's always better 
to rerun too many components, than too few, whenever anything is changed.

>
> - HMMVoiceMakeVoice: The headers of pdf file problem was because Mary
> 4.3.0, along with hts 2.1, hdecode 3.4 and htk 3.4, works only with
> hts-engine 1.01. Solution: download and install it.

So it says on 
http://mary.opendfki.de/wiki/HMMVoiceCreation#IICheckthenecessaryprogramsandfiles:

>
> In order to gain time for solving the problem, I created a test-voice,
> with only 50/3000 sentences. Of course the result is not good, and I will
> also have to work with the syllabifier, but at least, this is the way.
> I've attached a short demo.

I'm curious to hear some output... But don't underestimate the 
importance of fixing the syllabification to improve results. Also, in 
your current setup, I would expect bizarre errors for very long words.

Best wishes,

-Ingmar

>
> Anyway, the real voice, with the 3000 sentences, is being aligned by the
> EHMM aligner. We will definitely have it tomorrow.
>
>
>
> Will send you the real demo tomorrow,
> Florent
>
>
>
> _______________________________________________
> Mary-users mailing list
> Mary-users at dfki.de
> http://www.dfki.de/mailman/cgi-bin/listinfo/mary-users

-- 
Ingmar Steiner
Postdoctoral Researcher

LORIA Speech Group, Nancy, France
National Institute for Research in
Computer Science and Control (INRIA)