To respond to my own ticket (#181), I patched dsp.c to disconnect on 1
minute of silence.
It occurs to me that the ticket is more of a discussion. So here I am in
the dev list.
The patch to disconnect on silence turned out to be relatively easy but it
still leaves a lot to be desired. I also tried to document what I was
learning about dsp.c while I was at it.
The only thing my patch does is read the value of dsp->totalsilence and
disconnect when > 60000. Of course it took me hours to figure this out...
Sigh.
It rests on an assumption about dsp->totalsilence. Here is the formal
question:
"What does a value in dsp->totalsilence represent?"
And here is my attempt to answer it:
int totalsilence /* duration of current silence in ms. 0 means the
current frame was not silence. 20 means that the current 20ms frame was
the first detected silence. Negative values are undefined. */
To integrate this into callweaver it seems there needs to be a way to
configure this "feature" from a configuration file such as zapata.conf.
How would I make that happen? Should it piggy back on busydetect for
example? At present my test occurs inside
#ifdef BUSYDETECT
int opbx_dsp_busydetect(struct opbx_dsp *dsp)
Perhaps a new option "detectsilencedisconnectseconds=<number>", where the
default detectsilencedisconnectseconds=0 means don't disconnect on
silence.
I also noted that opbx_dsp_silence would do real harm due to an
unintialized value for len if the audio stream is not OPBX_FORMAT_SLINEAR
(callweaver-RC-1.1.99.20070815).
The other thing that my patch does which I don't like is test totalsilence
every frame. What a waste of CPU time. How can I reduce the frequency?
Surely silence detection and busy detection doesn't need to happen every
frame. How can the granularity be adjusted? Perhaps once every 500 ms.
The next thing I want to implement is long dial-tone detection disconnect,
because this is the only option I have available with the traditional PBX
I'm integrating with.
And then from a larger perspective, isn't busy detection, silence
detection and long dial-tone detection a sub-feature of call-progress
detection...
I found the call-progress code impenetrable - I also found that it doesn't
work, as you can read in ticket 181. Any hints?
I also wasted considerable time tinkering with busydetect options until I
read the code and realized that the values I was using were completely
ignored in callweaver-RC-1.1.99.20070815.
I added some LOG_WARNING messages as follows in the hopes that it will
reduce the amount of time wasted by fools like me:
void opbx_dsp_set_busy_count(struct opbx_dsp *dsp, int cadences)
{
/* 2007Aug23 ojw Oliver Wilcock: increase transparency through warnings
and debug messages
*/
opbx_log(LOG_DEBUG, "dsp busy count requested: %d\n", cadences);
if (cadences < 4) {
cadences = 4;
opbx_log(LOG_WARNING, "dsp busy count increased to hard coded
mimimum of %d\n", cadences);
} else if (cadences > DSP_HISTORY) {
cadences = DSP_HISTORY; /* ojw: I wonder if this should be
DSP_HISTORY/2 */
opbx_log(LOG_WARNING, "dsp busy count decreased to maximum history
window of %d\n", cadences);
}
dsp->busycount = cadences;
opbx_log(LOG_DEBUG, "dsp busy count set to %d\n", cadences);
}
void opbx_dsp_set_busy_pattern(struct opbx_dsp *dsp, int tonelength, int
quietlength)
{
#ifdef BUSYDETECT
opbx_log(LOG_WARNING, "dsp busy pattern ignored when using #define
BUSYDETECT\n");
#else
dsp->busy_tonelength = tonelength;
dsp->busy_quietlength = quietlength;
opbx_log(LOG_DEBUG, "dsp busy pattern set to %d,%d\n", tonelength,
quietlength);
#endif
}
You can see from my comment that I don't understand what DSP_HISTORY is.
It affects the length of some arrays but why would the maximum value for
cadences be DSP_HISTORY? I understand that there needs to be a
relationship but wouldn't it be something like DSP_HISTORY*mystery
unit/(busytone duration in ms + silence duration in ms)?
_______________________________________________
Callweaver-dev mailing list
[email protected]
http://lists.callweaver.org/mailman/listinfo/callweaver-dev
