[hts-users:01964] Re: Polish hts voice - question and win file problems

[Geoffrey Wilfart <geoffrey.wilfart@xxxxxxxxx>]

Hi,

The risk is if you have phones like, e.g. t^s and t.

In that case, question pattern "t^s^*" will obviously also match "t^*", which may lead your phone t^s to be wrongly classified.

I believe that as a generic rule, you should adopt a representation of the phones that causes no ambiguity on the labels.

I had the same issue, and chose to replace special characters by other strings (2-letter strings in my case).

Best is probably to replace your phones in the .utt files, so that I think you will be able to use the demo tools through the rest of your training process (labels generation, questions, etc.)

For test, you might need to modify also tools that will generate the labels before synthesis.

The *.win files are used to generate derivatives of your coefficients through the window.pl script. The first value (integer) indicates the number of coefficients to read, i.e. the window length.

The coefficients are weights to the parameters.

For instance, the mgc.win2 file extracts deltas from MGC coefficients. It states: 

3 -0.5 0.0 0.5

meaning that d[t] = -0.5*x[t-1] + 0.5*x[t+1]

where x[t] is the value of the coefficient at time t, and d[t] the corresponding delta.

Hope that helps,

Geoffrey

2009/4/30 Tomasz Kuczmarski <faqster@xxxxxxxxx>

Hello everybody,

I have already prepared a Polish clunits voice which works fine (mainly thanks to good labelling quality).

Now I am getting ready to train a hts voice from the resulting database.
So far I only have *.utt and *.raw files ready.

Now, I am trying to modify the question file for Polish.
In the phoneset of the clunits voice phones like "t^s, t^S, d^z, d^Z, t^s+" appear.
I have noticed that the syntax of the question file uses caret "^" as a special character.
The question obviously is: would there be any problems or conflicts because of the caret appearing in my phoneset and therefore having lines like
 
QS "LL-Fricative"            {t^S^*, ...}

in my question file?


Another question is about the   *.win files
Could somebody briefly explain the syntax and the meaning of the values of those files, please?
I have been searching for information on win files in the htk_book as well as on the hts-users ML but so far found no answer.


Any help would be greatly appreciated!



--
Best regards,

Tomasz Kuczmarski



e-mail: faqster@xxxxxxxxx
skype : faqster