[Speex-dev] Adjustable parameters for VAD in preprocessor

Jean-Marc Valin Jean-Marc.Valin at USherbrooke.ca
Sat Sep 24 03:18:08 PDT 2005


Hi,

If you can come up with a clean (non-intrusive) patch, I'm willing to
apply it. Ideally, it should go through int parameters to
speex_preprocess_ctl(). I think someone sent a similar patch to the list
a while ago, but I didn't have time to do the little cleanup it needed.

	Jean-Marc

> Hi,
> 
> Would a patch to change the constants on line 454 - 462 in the preprocessor 
> into variables be of general interest? At the moment, whether or not 
> "is_speech" is 1 is hardcoded to be (speec_prob > .35 or >.1 within last 20 
> frames or it's been less than 20 frames without).. I'd like to turn those 4 
> constants into speex_preprocess_ctl tunable variables, most likely through 
> SPEEX_PREPROCESS_GET_VAD_PARAMETERS / SET_VAD_PARAMETERS with ptr being a 
> pointer to
> 
> typedef struct {
>    float trigger_level;
>    float keep_level;
>    int keep_window;
>    int tail_length;
> } SpeexPreprocessVadParameters;
> 
> I'm a bit unsure if keep_window (used with prob > .1) should be the same as 
> tail_length. At the moment, I have an external counter in my program that keeps 
> track of how many frames it's been since the vad returned 1, and if it's less 
> than a user-set ammount, treat it as if it's still voice. This works well for 
> users that have slightly long pauses between sentences or words.
> _______________________________________________
> Speex-dev mailing list
> Speex-dev at xiph.org
> http://lists.xiph.org/mailman/listinfo/speex-dev
> 
-- 
Jean-Marc Valin <Jean-Marc.Valin at USherbrooke.ca>
Université de Sherbrooke
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://lists.xiph.org/pipermail/speex-dev/attachments/20050924/2431798e/attachment.pgp


More information about the Speex-dev mailing list