[Speex-dev] Proposed AGC additions

[Steve Kann]

Tom,

Are you also using the VAD from the preprocessor?  Does the VAD
detect this noise as speech?

I've got this same thing out through a lot of testing (via
iaxclient.sf.net), and I've never seen what you've reported here, and
I'm wondering if it is because I'm using VAD..

-SteveK

tgrand at canvaslink.com wrote:

>I've been using Speex's AGC in my VoIP program with mixed
>results.  One of the problems I had was:  At the beginning
>of the call, if the user hadn't spoken yet and there weren't
>any background sounds for the AGC to lock on to, most of the
>time the AGC would almost immediately jump way up to 200x gain
>(the default max).  This is undesirable and usually results in
>nasty feedback.
>
>If there ARE any background sounds occurring before this AGC
>jump, and they survive the noise filter, then AGC behaves
>nicely and hangs out around 1.33x to start with, gradually
>rising as it listens for something useful.  For example,
>as I tested this today, the faint sound of birds chirping
>outside was enough to prevent the AGC from spazzing out!
>
>Anyway, what I propose are the following commands to let
>an application interface a little more with the AGC:
>
>SPEEX_PREPROCESS_GET_AGC_GAIN
>SPEEX_PREPROCESS_GET_AGC_MAX_GAIN
>SPEEX_PREPROCESS_SET_AGC_MAX_GAIN
>
>I've implemented these and use them for the following:
>
>1) I display the current AGC gain to the user.
>2) If the AGC gain has reached its maximum, I don't transmit
>any audio.  This condition indicates to me that the AGC has
>not found any usable signal and it's very likely that the
>result of such extreme amplification is unwanted noise.  Once
>the user says something, the gain will back off and I can
>resume transmission.
>
>In the future, I might adjust the max AGC gain based on the
>results of a "mic setup wizard" type of thing, and/or the
>highest gain reached (and sustained) during the previous call.
>So that's where the SPEEX_PREPROCESS_SET_AGC_MAX_GAIN would
>come in.  I don't think there's any point in trying to do a
>SPEEX_PREPROCESS_SET_AGC_GAIN so I didn't list it here,
>but if anyone can think of a reason please say so.
>
>It's easy to implement these commands:
>
>In preprocess.h, SpeexPreprocessState:
>   float  agc_gain;
>   float  agc_max_gain;
>
>In preprocess.h, #defines at the end:
>#define SPEEX_PREPROCESS_GET_AGC_GAIN 14
>#define SPEEX_PREPROCESS_SET_AGC_MAX_GAIN 15
>#define SPEEX_PREPROCESS_GET_AGC_MAX_GAIN 16
>
>In preprocess.c, speex_preprocess_state_init():
>   st->agc_gain = 1;
>   st->agc_max_gain = 200;
>
>In preprocess.c, speex_compute_agc(), before last for loop:
>   if (agc_gain > st->agc_max_gain)
>      agc_gain = st->agc_max_gain;
>   st->agc_gain = agc_gain;
>
>In preprocess.c, speex_preprocess_ctl():
>   case SPEEX_PREPROCESS_GET_AGC_GAIN:
>      (*(float*)ptr) = st->agc_gain;
>      break;
>   case SPEEX_PREPROCESS_GET_AGC_MAX_GAIN:
>      (*(float*)ptr) = st->agc_max_gain;
>      break;
>   case SPEEX_PREPROCESS_SET_AGC_MAX_GAIN:
>      st->agc_max_gain = (*(float*)ptr);
>      break;
>
>And that should do it.
>
>Tom
>
>P.S. That problem I was having the other day with state being
>retained was my own fault, of course. :)  Speex was working
>just fine!  Sorry for wasting time on the list with it.
>
>_______________________________________________
>Speex-dev mailing list
>Speex-dev at xiph.org
>http://lists.xiph.org/mailman/listinfo/speex-dev
>
>
>