message from "Kyrian (List)" <[email protected]> to festival-talk
= = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
Hi Folks,
Well, I'm still fighting all this software to make a diphone voice of my
own.
I thought I'd done something wrong somewhere along the lines with my
label files (lab/*.lab) coming out wrong, and had spent a lot of time
manually labelling the end results.
However today I got a bit sick of it and went back to first principles
to try and work out what's going on, because the documentation strongly
suggests that automated labelling is better than spending the time doing
it manually for the majority of cases.
So, back I went, I thought I'd seen something odd in the make_labs code
where it chops off the last line of the label file with ch_track,
because it might be chopping off a pause instead of a 'real' diphone.
However that made things worse, so I guess I was barking up the wrong tree.
Anyway, I eventually checked out my prompt labels (prompt-lab/*.lab),
and found at least the first one I got to was quite badly broken, and in
a similar way to how my final labels (lab/*.lab) were being wrongly
generated as well, so I've reasoned (granted, however, I took the first
one of the prompt label files to indicate a trend without checking
further, which may have been premature) I think logically that if the
prompts are labelled wrongly, my final label files can never be correct
as they are based on them? I tried the steps to rebuild the prompts from
a source text file, but that still had the same error.
My questions, therefore, to myself and the list at large, are:
- How did the prompt labels get so badly broken?
- Do the prompt labels usually require such manual intervention?
- How can I fix them?
- Is it more likely I did something wrong in the initial steps to make
this happen?
The pattern, if there is one, of these prompt labels and final labels
being broken is that the labels are for shorter durations than the
diphones in the recorded .wav files, and therefore they tend to get out
of sync something like 1/3-1/2 the way through the recording, resulting
in a huge 'pau' at the end of the label file which encompasses the last
couple of diphones.
Any assistance that people can give would be greatly appreciated.
Yours,
Kev Green.
--
Kev Green, aka Kyrian. E: kyrian@ore.org WWW: http://kyrian.ore.org/
Linux/Security Contractor/LAMP Coder/ISP, via http://www.orenet.co.uk/
DJ via http://www.hellnoise.co.uk/
Human Rights left unattended may be Removed, or Destroyed, or Damaged
by the Security Services.
= = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
= University of Edinburgh's Festival Speech Synthesis System =
= http://festvox.org/festival Sent Via [email protected] =
= To unsubscribe mail [email protected] =
= = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
_______________________________________________
Festlang-talk mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/festlang-talk