[asterisk-users] Issues with Vestec ASR

Richard Kenner kenner at gnat.com
Tue Jun 8 13:35:56 CDT 2010


> Make sure that you only have the "one" grammar active when doing your test.
> You want the voice engine to basically only have 11 possibilities to chew on
> (0-9 plus "oh").

I always only load one grammar.  In the test I did below, there were
exactly TWO possibilities:

> I'm having a lot of problem with it recognizing "oh" for zero.
> 
> I've tried both "o" and "oh".  In one case, I just tried:
> 
> 	$digit = o { out = "0"; } | fundamental {out = "2"; };
> 
> So I gave it a choice that was VERY far away from what I said.
> 
> But when I said "o o o o o", more than 75% of the time, it had a bunch
> of "fundamentals" in its output.
> 
> E.g.:
> 
> 23:38:00>>      *Process Audio
>                         audio len: 2.17 sec (17384 samp)
>                         cpu sec: 0.038
>                         num sent: 1
>                         o fundamental o o fundamental (zipcode: 5 849 40 3
> 55)
> 
> I set vasre.conf to
> 
> -ns 3
> -nb 3
> -ac 10000
> -rj 40
> 
> which should make it easier to recognize the short word.
> 
> I emailed Vestec once but it was a black hole.
> 
> Does anybody have any ideas here?



More information about the asterisk-users mailing list