Speaker verification prototype ready

["Padraig O'Briain" <Padraig Obriain Sun COM>]

Cross-posting to Gnome Accessibility List as some members of that list may be 
interested.

Padraig

------------- Begin Forwarded Message -------------

To: Gnome Devel List <gnome-devel-list gnome org>
Subject: Speaker verification prototype ready

This is a follow-up for speaker verfication in gdm that was discussed a while
ago. I have just put a (static) binary release of my speaker verification
software here:
ftp://freespeech.sourceforge.net/pub/freespeech/static_SV.tgz

The tarball contains the following files:
sv_sound  : The statically-linked linux/x86 executable
si.gmm    : A speaker-independent voice model
jm.gmm    : The speaker-dependent model for my (Jean-Marc) own voice

To test it, all you have to do is:
% sv_sound jm.gmm
 and then speak (for a couple seconds) in the microphone. The recording will
stop after one second of silence. The result will then be printed. (note that
"si.gmm" needs to be in the current directory)

You will get something like 
<Float 3.50685 >
The number represents the likeliness of your voice being that represented my the
model in argument (in this case, my voice). Usually a number above 10 means that
the model matches the voice, while below 10 it doesn't match. Of course it's not
perfect, but I hope to improve accuracy over time.

I did not provide source code at this time because it is automatically generated
with the Overflow (http://freespeech.sourceforge.net/overflow.html) environment.
You can download both Overflow and the XML file used to generate the code from
http://sourceforge.net/project/showfiles.php?group_id=590

I'm currently working on two (rather easy) improvements:

1) Building as a library so it can be linked from a program (like gdm)
2) Having speaker *Identification*, that is you speak and the system (gdm, kdm)
logs in the right person without needing to enter your login.

For those who (of course) would like to train their own voice model, you can do
that in two ways:
1) Get the "Overflow" and "Open Mind Speech" packages and read the instructions
2) record your voice at 16kHz/16bit/mono, upload it to the omspeech.dyndns.org
ftp site (go to incoming/<language>) and e-mail me the name of your file. I will
e-mail you back a .gmm model of your voice.

Have fun!

	Jean-Marc

P.S. The code you can download is a bit older that what I'm releasing now, the
newer stuff is only in CVS.

_______________________________________________
gnome-devel-list mailing list
gnome-devel-list gnome org
http://mail.gnome.org/mailman/listinfo/gnome-devel-list

------------- End Forwarded Message -------------