[Speex-dev] preprocesssor questions

[Jean-Marc Valin]

Eran Hashin a écrit :
> Hi
> Thank you for your work.
> We are trying to use the speex SW with a C6000 TI DSP.
> I am working with an HW codec that does AGC by HW inside.
> Will doing the AGC before the preprocessor make a problem?

Yes, expect weird behaviour when the AGC gain changes.

> Does VAD work well in a noisy eviroment?

The current VAD isn't very good. The one in 1.2beta1 was a bit better,
but not great either.

> In SW there are some remarks on things to be fixed, can I count on SW as working or are some part still to be improved.
> I also tried to understand SW but I guess I was unable to.
> Where are the version will the latest patches located?

You mean the git/svn repositories?
http://git.xiph.org/speex.git
http://svn.xiph.org/trunk/speex

	Jean-Marc

> Best regards,
> Eran Hashin
> 
> 
> -----Original Message-----
> From: speex-dev-bounces at xiph.org [mailto:speex-dev-bounces at xiph.org]On Behalf Of Jean-Marc Valin
> Sent: Saturday, May 17, 2008 10:58 AM
> To: florian.winterstein at gmail.com
> Cc: speex-dev at xiph.org
> Subject: Re: [Speex-dev] preprocesssor questions
> 
> 
> Florian a écrit :
>> 1) I looked at the source code. Even though I am an experienced
>> programmer it looks like Chinese to me ;-) What would I have to do to
>> modify / tweak the audio processing code for my needs? Are there
>> documents describing the algorithms applied for AGC, NS and VAD or does
>> one have to understand the source code? How do people help you modifying
>> the library  - do they magically understand the code?
> 
> I think normally it helps when you actually understand what you're
> changing. The other day, I was playing around with the Linux kernel
> source code and removed the scheduler. For some reason, the new kernel
> failed to boot. Strange...
> 
>> 2) I run the preprocessor it seems to take the same amount of time to
>> run VAD+NS+AGC as if I only run VAD. Is it possible to restructure the
>> code in a way that saves cpu cycles if I only run the VAD?
> 
> Yes.
> 
>> 3) Or more specifically is it possible to only run the VAD and only if
>> the VAD returns speech do the real precessing (AGC, NS). Basically save
>> CPU if there is no real voice.
> 
> No, or else the AGC and NS will suck.
> 
>> 4) You are writing that people are working on improving the VAD. Again
>> are there documents decribing the changes or how does one get started?
> 
> No, sorry. The little time I have to spend on this is spent on the
> actual code rather than doc.
> 
> 	Jean-Marc
> _______________________________________________
> Speex-dev mailing list
> Speex-dev at xiph.org
> http://lists.xiph.org/mailman/listinfo/speex-dev
> 
> ______________________________________________________________________
> This email has been scanned by the MessageLabs Email Security System.
> For more information please visit http://www.messagelabs.com/email 
> ______________________________________________________________________
> 
> 
> ______________________________________________________________________
> This email has been scanned by the MessageLabs Email Security System.
> For more information please visit http://www.messagelabs.com/email 
> ______________________________________________________________________
> 
>