I've reviewed earlier posts about completetimeout not being supported. Can someone provide the latest status of this property relating to CVP v11?
1) Is completetimeout supported in later versions of CVP like v11?
2) If not supported what options are available to provide this property in a recognition element request?
3) Is there some explanation why CVP will not support this property when the other IVR vendors do provide the support?
Appreciate any information you can provide.
I can't answer all your questions (I'm not Cisco employee) - but it was
always that Nuance didn't support the completetimeout. It wasn't a Cisco
issue. Nuance does support incompletetimeout. Do you know if Nuance 10
supports completetimeout now?
We're actually testing the MRCP-v2 protocol across multiple IVR vendors. As we monitor the payload between from Cisco to the ASR (MRCP-v2) we don't see the payload including the "completetimeout" even though the property is clearly set in the application. While monitoring other IVR vendors we do receive the property.
Are you aware of any options to force CVP to include the property, or can the property be passed in any other fashion? We're testing against CVP v11.
Rajen, Can you weigh in on the question below? Is there a way to have
Cisco pass the completetimeout to Nuance?
Some additional details that may be useful and perhaps trigger a work-around solution. We're interacting with a complete language model ASR (Natural Language). Our test scenario is a caller speaking a mixed list of digits, and numbers like
"1 2 A E 3 3 F G R T X JH". By default the ASR detects 1/2 second of silence as the caller has stopped speaking and completes the recognition. Obviously, as a person is reading the expected input they may falter, and easily expose the 1/2 second silence detection so we need to allow the caller more time as they read the letters, and digits. To accomplish this additional time we set "completetimeout" and "incompletetimeout" to 3 seconds each.
These setting give the caller a nice window of opportunity to speak the response at a natural speed, and not encounter a recognition complete in the middle of the string.
Testing with a non-Cisco IVR this approach works perfectly, and we see in the MRCP traces that both of the time outs are passed in the MRCP exchange. These time out settings combined allow the caller more time between speaking the letters, and numbers and not be interrupted by a recognition complete.
Executing the same test scenario on CVP we see that the "completetimeout" is not forwarded in the MRCP interaction so the ASR is not changing it's behavior and only allowing the caller 1/2 second silence window before detecting recognition complete.
Our primary goal is to have the "completetimeout" passed over the MRCP-v2 session so it can be used by the ASR.
With Cisco if you set the incompletetimeout to 3s then it'll be 3s
timeout between EACH word uttered before assuming the caller is done
speaking. So that they use the incompletetimeout between words AND to
know the caller is done.
I just tried this with a digits grammar. So I could say "one" and after
3s the system would tell me I said "one"
or I could say "one (2s pause) two ((2s pause) three (2s pause) four (3s
pause)" and the system would only move on once it encountered the 3s pause.
Just to clarify we are interfacing with a non-nuance ASR so our testing against other IVR platforms has proven that this alternate ASR uses both incompletetimeout, and completetimeout to control the time between utterances, and end of speech detection.
I did try setting only incompletetimeout = 3s but had no change in testing results. As the list of letters and numbers was spoken the caller is cut off. The only way to pass the recognition is to speak very quickly.
Appreciate any ideas on how we can cause complete timeout to be passed as part of the ASR payload.
Completetimeout only pertains to collecting speech when the caller has
said the maximum number of valid digits.
From the VXML Spec'n:
/The completetimeout is used when the speech is a complete match of an
active grammar. //
/By contrast, the incomplete timeout is used when the speech is an
incomplete match to an active grammar./