We have a speech reco application (using the Microsoft.Speech SDK included with UCMA SDK install). Our first shot at implementing barge in was to start the recognizer and start the synthesizer at the same time and when we get a speech detected event from
the recognizer we stop the synthesizer. That all works fine. The issue is the intial silence timeout is set to 3 seconds so after three seconds from when the prompt start playing, we get a end recognition event (the initialSilenceTimeout fired). The prompt
is still playing (say it it 10 seconds long), so we cancel it. The caller never had a chance to say anything in this scenario. When we wrote this application on OCS Speech Server, the API was 'smart enough' to somehow tie the synthesizer and recognizer together
so that even if you started the synthesizer and recognizer at the same time, the recognizer's initialSilenceTimeout timer would not start running until the prompt stopped playing. That's exactly the functionality we are trying to duplicate here. We need to
start both at the same time, but we need the recognizer's initial silence timer to start only after the prompt has finished playing
The solutions that come to mind are to initially set the recognizer's initialTimeout to a very long time and then if we ever get the synthezier completed event, we try to set the timeout to the normal 3 second value we want. This does not appear to work.
Once the recognizer is running, I am pretty sure you can set the intialSilenceTimeout property.
Next possible solution is to start the recognizer and synthezier like we do now, and we would have to have the initialSilenceTimeout value set to a high value. Then if the prompt ever ends, and we haven't had any speech yet, we stop the recognizer and restart
it using the 3 sencd tiomeout we want. This solution worries me because there might be a window between restarting the recognizer where we miss some speech and that might affect recognition.
So the first solution probably just doesn't work, and the second is pretty ugly. It seems like there must be a way to do this? I really think there is a better way to do this, but searching through the class library reference, I just can't see how. Anyone
have any ideas or experience trying to do this?
When I search for barge-in in the Microsoft Speech SDK help doc, it talks about something that might apply but it is using the
Microsoft Speech Platform API which is a C++ CPI that we aren't using. It seems like
Sorry if this not so much a ucma question as a Microsoft.speech question. I couldn't find a forum for Microsoft.speech (just for OCS speech server). If you know of another forum that might be more applicable, please let me know.
Thank you
mike castillo
