Speech-to-text initialization hanging

Ivan H 1

Hi, I'm using speech-to-text services C++ SDK for Linux. I've build a test app following the guidelines from here:
https://video2.skills-academy.com/en-us/azure/cognitive-services/speech-service/get-started-speech-to-text?tabs=windowsinstall&pivots=programming-language-cpp
It works fine when subscription key is accepted but when there is some issue (e.g. error code 4 about running out of quota) it takes quite a long time (up to 10 seconds) to receive a response and my app freezes. If I move the logic to a separate thread it still causes freezing on the app exit and I receive an error code 6 ("Timeout while waiting for service to stop"). What is worse - sometimes I don't receive any response at all and therefore I need to force terminate my app as it sticks completely.
Is there any ability to reduce this delay? Could you please advise?

romungi-MSFT 44,771 Reputation points Microsoft Employee

2021-09-16T14:11:19.44+00:00

@Ivan H Are you using the free tier for your speech resource? Free tier does have a limit that limits the TPS and the amount of audio and text that can be processed. The quotas and limits page does provide some guidance on some of the quota issues.

For the issue with your app where it needs force terminated I would recommend to check if the same occurs when you change your tier to standard. This should help identify if the issue is with the quota or any dependencies that are required for your app to work on linux.
Ivan H 1 Reputation point

2021-09-17T05:41:07.307+00:00

Thanks @romungi-MSFT
But TBH I expected to receive an error message (about limitation) without waiting ~10 seconds. Or, even better, to have an ability to terminate initializing request at any time.
Basically, my question is about the continuous recognition. When should I call this:

recognitionEnd.get_future().get();

Just after StartContinuousRecognitionAsync? Should it be executed in a separate thread?
romungi-MSFT 44,771 Reputation points Microsoft Employee

2021-09-17T14:47:04.503+00:00

@Ivan H With continuous recognition the call to StopContinuousRecognitionAsync() is required to stop the recognition. I am not sure if the recognition in your case is taking 10 seconds though, depending on the input and audioconfig used.
The call to recognitionEnd.get_future().get(); is after the recognizer->StartContinuousRecognitionAsync().get(); call in the same thread. This sample should help you run and understand the scenario better.
Ivan H 1 Reputation point

2021-09-20T06:45:23.103+00:00

OK @romungi-MSFT , the example is about recognition from a file. What about recognition from a microphone? I tried adjusting the example to recognize from microphone and I cannot stop it, perhaps it waits for some timeout.
I can see there an example of microphone recognition but it is not continuous.
BTW, is it possible to call stop recognition from another method of the class?
romungi-MSFT 44,771 Reputation points Microsoft Employee

2021-09-21T10:56:04.177+00:00

For continuous recognition scenarios with mic you would like to configure an option like a button event to start and stop the recognition. This issue kind of provides a high level guidance on this usage for C#. In this case you should be able to call the stop event if your recognizer is global.

Share via

Speech-to-text initialization hanging

Your answer