[Asterisk-Dev] Voice detection

[Steven Critchfield]

On Fri, 2003-10-03 at 00:08, Brad Waite wrote:
> Steven Critchfield wrote:
> 
> > On Thu, 2003-10-02 at 21:36, Brad Waite wrote:
> > 
> >>Does anyone know if there's public voice detection algorithms available?  I've 
> >>scoured the net for the last hour or so, and I can't come up with anything 
> >>except a few proprietary or embedded solutions.
> >>
> >>I know dsp.c uses goertzel algorithms for DTMF detection, but how does one 
> >>detect voice?
> >>
> >>I dunno, maybe detecting voice isn't the way to go.  I want to begin playback of 
> >>a file after a phone/answering machine has answered.
> > 
> > 
> > Argghh telemarketing....
> 
> Argghh voice mail notification...  (among other non-telemarketing things)  :)

I wonder if for all you who thinks your voice mail is so important that
you can't have it wait till you normally would check(not just you Brad),
if you wouldn't be better served with dynamic forwarding of a "on-call"
phone. 

I understand that your lives are different from mine, but if it is so
important that it has to be dealt with immediately, there is a number in
our office that forwards to the on-call person. At that point we can
answer the line wherever we have forwarded it too. Of course we also
tend to have our customers email us if it wasn't so important as to need
immediate attention.

Other than your voice mail notification, the only other uses I know of
are essentially telemarketing of nature, either soliciting of any kind
or bill collectors. I even lump the local politicians into the solicitor
category. I hated seeing that 2 of my local city council candidates used
a automated caller to solicit my vote this last 2 elections. 

> > shouldn't you detect silence and then start?
> 
> Possibly, but I would think that between the last ring and the speech there 
> could be enough silence to return a false positive.

Essentially what you need to do is be on a digital line so that answer
is definitively known, and then what till you detect non silence, and
then wait till you detect approximately half to a 1 second of silence.
For digital voice recording devices, this should be ample time. For any
of the older tape based systems, you might start too early. 

Silence detection is already a function in dsp.c. In fact, I was very
surprised to see it even in the RECORD function of AGI now.



-- 
Steven Critchfield <critch at basesys.com>