[hts-users:00297] Re: training a new voice

[donall.mccann5@xxxxxxxxxxx]

I just realised that the f0 error was because of the headers in my f0 files.
I have since corrected that (by removing them entirely) and I now get a new
error

convert f0/arctic_a0001.f0 to log F0
arctic_a0001.lf0
Can't take log of -3.68935e+19 at scripts/freq2lfreq.pl line 53.

I suppose this means that my values are singificantly different (or in a
different format) to the demo ones. Were the f0 values calculated using the
festvox get_f0 script? If not, what should I be using?

Donall


>-- Original Message --
>Date: Fri, 12 May 2006 21:11:57 +0100
>From: donall.mccann5@xxxxxxxxxxx
>Reply-To: hts-users@xxxxxxxxxxxxxxxxxxxxxxxxx
>Subject: [hts-users:00296] training a new voice
>To: hts-users@xxxxxxxxxxxxxxxxxxxxxxxxx
>
>
>Hi. I'm trying to use the training script (for Arctic SLT) supplied with
>2.0 to train a different voice by simply supplying it with different data.
>I've included different raw, f0, mcep, lab and utt files trained using festvox,
>but the training script is giving me some errors. They are as follows:
>
>"convert f0/arctic_a0001.f0 to log F0
>arctic_a0001.lf0
>Invalid type '.' in unpack at scripts/freq2lfreq.pl line 47."
>
>and
>
>"# generate fullcntext model list occurred in the training data
>basename: too many arguments
>Try `basename --help' for more information.
>basename: too many arguments
>Try `basename --help' for more information.
>/bin/sh: line 0: [: too many arguments
>sort: open failed: tmp: No such file or directory"
>
>The rest of the script seems to run fine and it is generating the appropriate
>files (or, at least, they look like the appropriate files!).
>
>I ran the script successfully with the original data and so I don't think
>it should be a code bug. Is there any data I haven't supplied or is there
>anything that I supplied in the festvox format that should be in a different
>format?
>
>Donall
>
>