[asterisk-speech-rec] Re: Beta4 causing speech rec performance changes?

[Tom [toms400]]

I went back and tried beta3 and I have the problem there, as well.
Sorry to point the finger at beta4. It seemed to be more pronounced
and I've gotten more complaints since I switched.

I've got vad_burst_threshold at 200 ms, vad_noise_floor at 12000,
barge-in at 100, eos delay at 1000ms and end_of_speech_timeout at 8000
ms and I'm still having the problem. If I set vad_noise_floor any
higher, I can't detect speech of a comfortable volume.

The tone blips range between 6 ms and 20 ms in length and, as far as I
know, are the result of partially clamped dtmf tones (which I get from
every SIP gateway provider I've tried).  I get them with every cell
phone and landline phone I've tried.

Ideally, I'd love to get rid of the tone blips entirely as they're
also a problem during conferences. If you know how to do this, please
contact me off-list.  If this isn't possible then perhaps a
modification could be made so that once out-of-band dtmf is detected.
the speech app only listens for more out-of-band dtmf (no speech)
until a timeout or stop digit is reached?

I can provide wav files of the blips if that's helpful.

I'd appreciate any suggestions.
Thanks
Tom


On 12/25/06, Tom [toms400] <toms400 at gmail.com> wrote:
> A couple of days ago I installed 1.4 Beta4 and I noticed some changes
> in results SpeechBackground returns.  When users enter DTMF instead of
> speech, the engine seems more sensitive and instead of waiting for a
> concluding #, is exiting and returning a result too early.  I've
> checked the speech files and there is no background noise -- just some
> short tone blips.  I was not having this problem with Beta3.
>
> Any ideas?
>