Speex dev - Aug 2006 - agc

Hi,
 
I'm a new Speex list member, and so far I'm very impressed with Speex. 
BUT
automatic gain control seems to me to be virtually unusable.  If a recorded
utterance begins with even a tiny noise, that noise is amplified to the
maximum volume.  (See, e.g. the initial utterance in the TIMIT corpus, "She
kept your dark suit...")
 
This seems like a bug to me.  If list members agree, any idea on how to fix
it?
 
CB
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.xiph.org/pipermail/speex-dev/attachments/20060809/ddd66f5a/attachment.html

Le mercredi 09 ao?t 2006 ? 13:24 -0400, Christopher Bader a ?crit
:
> Hi,
>  
> I'm a new Speex list member, and so far I'm very impressed with
Speex.
> BUT automatic gain control seems to me to be virtually unusable.  If a
> recorded utterance begins with even a tiny noise, that noise is
> amplified to the maximum volume.  (See, e.g. the initial utterance in
> the TIMIT corpus, "She kept your dark suit...")
>  
> This seems like a bug to me.  If list members agree, any idea on how
> to fix it?
This is an inherent problem with any AGC. Before someone actually 
speaks, there's no way to know what level to use, so the AGC just 
"guesses" that noise is speech. The only way to fix it is with 
kludges and/or a priori knowledge.

	Jean-Marc