VoIP Mailing List Archives

[mszlazak at aol.com]

Session.collectInput looks like it has a fixed timeout setting which blocks the channel until the timeout expires.

I've been informed that appears to be the case from a brief examination of how speech is collected. SWITCH_DECLARE(switch_status_t) switch_ivr_collect_digits_callback
seems to use a fixed timeout to cancel recording. ??

Is it possible to set-up a "relative" timeout which is based on the time from when the last valid sound was recognized as in choosing toppings from a list of 20 toppings in the pizza demo? This situation involves an indeterminate number of valid sounds (toppings) and thus is unlike the dtmf example give in http://wiki.freeswitch.org/wiki/Session_collectInput where a counter in the callback is also used to determine when to unblock the channel.

Mark.


Instant access to the latest & most popular FREE games while you browse with the Games Toolbar - Download Now!

[anthony.minessale at g...]

The speech detector never stops working. The collect input just pauses to wait for some input.
It can catch the event any time a file is playing or that collect input is called.


On Sat, Nov 15, 2008 at 1:19 PM, <mszlazak@aol.com (mszlazak@aol.com)> wrote:

[mszlazak at aol.com]

I'll follow up more on this but before I do, is their a need to re-sample my audio for pocketsphinx?
My soft-phone is X-lite and the codec is ALC883.





-----Original Message-----
From: Anthony Minessale <anthony.minessale@gmail.com>
To: freeswitch-users@lists.freeswitch.org
Sent: Sun, 16 Nov 2008 8:40 am
Subject: Re: [Freeswitch-users] Relative timeout in Session.collectInput?

The speech detector never stops working. The collect input just pauses to wait for some input.
It can catch the event any time a file is playing or that collect input is called.


On Sat, Nov 15, 2008 at 1:19 PM, <mszlazak@aol.com (mszlazak@aol.com)> wrote:

[mszlazak at aol.com]

I'm suspecting that one of my pauses (or something else), as I'm saying my topping choices, is being interpreted as the end of my choices.

I've tried adjusting the timeout argument value of collectInput but that doesn't seem to do anything.

So, I'm left wondering what's causing my choices to be truncated and thus collectInput handles only a part of the toppings ordered.

It doesn't seem like the speech recognizer is only recognizing one topping at a time as an event that's caught by collectInput since I have gotten recognition of more than one topping but usually not that many more.

Nickolay Shmyrev at CMU Sphinx Help Forum thought there maybe some timeout involved because passing silence to the recognizer wasn't a good idea and suggested a relative timeout.

So what's being passed to the recognizer or is there something else going on?

Mark.

-----Original Message-----From: Anthony Minessale <anthony.minessale@gmail.com>

To: freeswitch-users@lists.freeswitch.org
Sent: Sun, 16 Nov 2008 8:40 am
Subject: Re: [Freeswitch-users] Relative timeout in Session.collectInput?

The speech detector never stops working. The collect input just pauses to wait for some input.
It can catch the event any time a file is playing or that collect input is called.


On Sat, Nov 15, 2008 at 1:19 PM, <mszlazak@aol.com (mszlazak@aol.com)> wrote:

[anthony.minessale at g...]

maybe its the detection of talk stop.
are you pasing between your toppings.
try saying them all very fast with no pause.
It may be the pause between utterances that is catching you.
the config has params.

threshold (default 400): higher the number louder you have to talk to be considered "talking"
silence-hits (default 35): number of hits below threshold before detecting "stop talking"

On Mon, Nov 17, 2008 at 4:23 PM, <mszlazak@aol.com (mszlazak@aol.com)> wrote:

[mszlazak at aol.com]

Yup, I tried different settings for dft_min and dft_confirm before but did not understand what they meant.
Back then I guessed that "dft" meant "discrete Fourier transform" not "default" and these variables were cut off frequencies of a filter but that did not quite make sense.
For instance, in the "yes/no" recognition part of the demo, one variable was set lower than the other but not in other places.

Thanks for clearing up their meanings.

I've played with values for those variables more based on the output and it does help.

However, I'm and was getting recognition of "strange patterns" that all start with 2222 in superscript like:

2222
2222||||?
2222h
2222p?o??
2222h\|
2222||||?

What's going on here and what can I do about it?

Thanks.
Mark.





-----Original Message-----
From: Anthony Minessale <anthony.minessale@gmail.com>
To: freeswitch-users@lists.freeswitch.org
Sent: Mon, 17 Nov 2008 2:39 pm
Subject: Re: [Freeswitch-users] Relative timeout in Session.collectInput?

maybe its the detection of talk stop.
are you pasing between your toppings.
try saying them all very fast with no pause.
It may be the pause between utterances that is catching you.
the config has params.

threshold (default 400): higher the number louder you have to talk to be considered "talking"
silence-hits (default 35): number of hits below threshold before detecting "stop talking"

On Mon, Nov 17, 2008 at 4:23 PM, <mszlazak@aol.com (mszlazak@aol.com)> wrote: