My recognizer stops after 30 seconds about cognitive-services-speech-sdk-js

[glharper]

@OmriBenShoham Thank you for using JS Speech SDK and writing this issue up. You asked about a 30 second limit using startContinuousRecognitionAsync()...that limit does not exist. Even with silence, continuous recognition should not stop until 10 minutes pass, or you call stopContinuousRecognitionAsync().

Note that we do have a react js integration sample using JS Speech SDK, but the sample uses the recognizeOnceAsync() API, meaning mic input stops on the first recognized result from the service.

I can adapt that sample to add continuous start/stop options...it wouldn't shock me if the ReactJS state primitives don't play well with ongoing event callbacks, but in theory continuous recognition should work with the sample. Could you look at that sample and adapt it first to see if it addresses this issue?

[OmriBenShoham]

@glharper Thank you for your fast response,
I was trying to adapt the example you sent me and use recognizer.recognizeOnceAsync
My recording was still stopping after 30 seconds.
Also I checked and i'm not calling stopContinuousRecognitionAsync.
Maybe there is a place where I can see the logs of the recognizer?

[glharper]

@OmriBenShoham I've added a branch with continuous recognition integration here. Let me know if it works for you.

[OmriBenShoham]

@glharper I was trying to use my recording which is 35 seconds long with the sample branch you provided.
It still cut me exactly after 30 seconds - is the Continuous working for you with a recording of more than 30 seconds?

[glharper]

@OmriBenShoham Can you try speaking into your mic for more than 30 seconds? The audio config for that branch is fromDefaultMicrophoneInput(). Alternatively, you could change the sttFromMicrophoneContinuous method to use file input (see the fileChange() method for the audioConfig setup needed.)

[OmriBenShoham]

@glharper I was speaking to my microphone for 35 seconds using "Convert speech to text from your mic (Continuous)." but only 30 seconds was applied.
I'm also adding image of the button I used from the sample branch you sent me

[glharper]

@OmniBenShoham For me, the results from microphone keep updating, even after 30 seconds. Don't think this is an SDK issue, or even an issue with the sample. To turn on Diagnostics, in the sttFromMicrophoneContinuous method, add this call at the beginning:

        speechsdk.Diagnostics.SetLoggingLevel(speechsdk.LogLevel.Debug);

If you can grab the console output and attach it here as a text file, I can take a look and see if I can understand what's happening.

[OmriBenShoham]

@glharper
Uploaded the console output logs as a file here.
I removed some token logs which was there(I guess it's not relevant) - I can see a InitialSilenceTimeout as a RecognitionStatus in some of the logs, but i'm not sure if it's related.
consoleOutputContinous.log

[glharper]

@OmriBenShoham I think your audio is cutting out on your end after 45 seconds or so. From the log:

• at 2023-10-24T15:01:40 audio recognition is started by service (turn.start)
• (multiple intermediate results (speech.hypothesis))
• 2023-10-24T15:02:21.185Z: first speech.phrase (you should see the first recognized text in the sample)
• (one more intermediate result (speech.hypothesis))
• 2023-10-24T15:02:24.445Z: the second (and last) speech.phrase (you should see the text change in your sample)
• 2023-10-24T15:02:44.165Z: initial silence timeout received, with no intermediate results

Something on your end is cutting off your audio at 45 seconds.

[OmriBenShoham]

@glharper
There is nothing special happening at my side after 45 seconds,
Can we maybe jump on a quick call / slack so I will be able to better demonstrate you the problem?

[glharper]

@glharper There is nothing special happening at my side after 45 seconds, Can we maybe jump on a quick call / slack so I will be able to better demonstrate you the problem?

Sure, send me an email at <my_username>(at)microsoft(dot)com and we'll set something up.

[OmriBenShoham]

@glharper
Thanks! Sent you an email

[OmriBenShoham]

@glharper
Kindly reminder :) sent you an email, if you can please reply and we will align a quick call

[glharper]

@glharper Kindly reminder :) sent you an email, if you can please reply and we will align a quick call

Not seeing it in my inbox, could you check where you sent it? (my username is glharper or GLHARPER in caps :-)

[OmriBenShoham]

@glharper Sent you a connection with note on Linkedin :)

[glharper]

Closing, resolved via Teams meeting

[GalDayan]

Hi @glharper
What are the scenarios when speechEndDetected is triggered?

[gerrieswart]

so what was the resolution for this?