[prev in list] [next in list] [prev in thread] [next in thread] 

List:       asterisk-dev
Subject:    Re: [Asterisk-Dev] Require Help in Detection of Human Voice or
From:       "Justin Newman" <jnewman () newmantelecom ! com>
Date:       2005-11-26 18:49:56
Message-ID: 093001c5f2ba$322e9680$fe14a8c0 () YODA
[Download RAW message or body]

> actually it can trigger 'talk' based on silence duration. (not just dtmf
> triggering)
>
> years ago i wrote the agentlogin doc series, final i did was
agentlogin3.rtf
> in that one there is a section that shows what *basic 'hello' detect
> would look like
> <>
> exten => 1,1,BackgroundDetect(nothing|500|300|1050) ; nothing is a 3 sec
> blank file
> ; the 500|300|1050 is a simple
> ; approach to detecting a<>
> ; "hello"

"Talk" detection is one basis for determining answer, however it's not
accurate and doesn't apply to human vs. machine detection. We haven't found
a solid way of doing this and do not anticipate finding one. Possibly a flag
will be passed in future signaling (SIP/IAX future addition:
IVR/machine/human/no-telemarketing flags?). This may be helpful in several
applications. For example, a callback or paging application may have no need
to sit on the line with IVR. It could quickly determine this and disconnect.

We have used a few a techniques to determine human vs. machine, such as the
presense of a beep, noise, repetition, and answer strength, among others.
When combined together, they work to some degree, however the all add
various problems such as answer latency and none of the techniques work on
large groups. As systems evolve, we run into another problem: the difference
between human and machine becomes less obvious.

Justin


[prev in list] [next in list] [prev in thread] [next in thread]