Rockbox mail archive
Subject: voice file generation (Re: english.voice using AT&T Natural Voices)
From: [IDC]Dragon (idc-dragon_at_gmx.de)
> Now appearing on my webspace ATT NV Crystal speaking english.lang v
> 1.109. I had to make a few minor corrections to the pronunciation due to
> some weirdness in their TTS engine but I've only done some quick testing
> on my recorder so there may be more!
Probably you mean "Resume" being spoken like curriculum vita?
> If there is interest in having their
> voice, Mike, speak it I also have that available. I'll try to keep up
> revisions to english.lang but I'm fatally lazy. :)
No real changes to the language file, it's not getting outdated so quickly.
> Anyway http://webpages.charter.net/ruiner1/english.voice is what you
> really want.
What parameters have you used to encode it? I had a look inside, half of the
file seems to be padding zeros. In general, you should use VBR to have more
efficient coding. And because of a bug in changing from one clip to another
it is currently better to disable the bit reservoir, Lame parameter --nores.
I have made a new program to generate the voice clips, see my tool
The new one is "lang2wav", which generates the whole bunch of speech clips
as .wav without a 3rd party program like TextAloud. Input is the .lang file,
it uses the default voice as configured in the Control Panel. After running
it, you can batch-encode (maybe first batch-trim) the lot and run "voicefont"
to make the final language file. Result should be not larger than 1.5 MB, else
tune the encoder parameters. So the process is pretty simple now, and
lang2wav is way faster than TextAloud.
We're still waiting for a volunteer to do something similar under Linux with
e.g. the "Festival" TTS program, hint, wink. This could then be part of the
+++ NEU bei GMX und erstmalig in Deutschland: TÜV-geprüfter Virenschutz +++
100% Virenerkennung nach Wildlist. Infos: http://www.gmx.net/virenschutz
Page was last modified "Jan 10 2012" The Rockbox Crew