【cognitive-services-speech-sdk】startContinuousRecognition will automatically stop

Question

【cognitive-services-speech-sdk】startContinuousRecognition will automatically stop

玉洁朱 0

I downloaded the official iOS SDK "MicrosoftCognitiveServicesSpeech-XCFramework-1.34.0.zip" from your official website.

I want to use [speechRecognizer startContinuousRecognition] to implement real-time continuous recognition, that the speech source come from microphone.

here is my relevant code:

AudioRecorder *recorder = [[AudioRecorder alloc]initWithPushStream:stream];
SPXAudioConfiguration *audioConfig = [[SPXAudioConfiguration alloc]initWithStreamInput:stream];

speechRecognizer = [[SPXSpeechRecognizer alloc] initWithSpeechConfiguration:speechConfig language:lang audioConfiguration:audioConfig];

[speechRecognizer startContinuousRecognition];

when I call [speechRecognizer startContinuousRecognition], it will begin recognizing, but will soon stop immediately.(even though i don't call [speechRecognizer stopContinuousRecognition] ) but if i sleep for N seconds, it will recognize N seconds, and then stop.

AudioRecorder *recorder = [[AudioRecorder alloc]initWithPushStream:stream];
SPXAudioConfiguration *audioConfig = [[SPXAudioConfiguration alloc]initWithStreamInput:stream];

speechRecognizer = [[SPXSpeechRecognizer alloc] initWithSpeechConfiguration:speechConfig language:lang audioConfiguration:audioConfig];

[speechRecognizer startContinuousRecognition];

[NSThread sleepForTimeInterval:3.0f]; // add this code will delay stop

What I expected

if i don't call [speechRecognizer stopContinuousRecognition];, I expect continuous recognition not stopping.

Rohit Mungi 49,131 Reputation points Microsoft Employee Moderator

2023-12-27T10:22:44.84+00:00

@玉洁朱 Could you try to check if this sample works for you?
玉洁朱 0 Reputation points

2023-12-27T11:43:55.4+00:00
@Rohit Mungi Thanks for reply.

That sample works for me. But the difference with my project is that:

the sample code's speech source is come from a file named "whatstheweatherlike.wav". But my project's speech source is come from microphone.

Since my speech source is come from microphone, I need use [recorder] to record sound, and then offer the stream to Azure speech recognition, my code is like:

SPXPushAudioInputStream *stream = [[SPXPushAudioInputStream alloc] init]; AudioRecorder *recorder = [[AudioRecorder alloc]initWithPushStream:stream]; SPXAudioConfiguration *audioConfig = [[SPXAudioConfiguration alloc]initWithStreamInput:stream]; [recorder record]; [speechRecognizer addRecognizingEventHandler: xxxxxxx [speechRecognizer addRecognizedEventHandler: xxxxxxxx [speechRecognizer addSessionStoppedEventHandler: xxxxx [speechRecognizer startContinuousRecognition];

When I run my code, it will begin recognizing, but will soon stop immediately, and i will receive a session stop event.

I compare my code with your sample code, and found that sample code has a while loop, and write data to stream untile the source file read the end. So I also add a while loop :

objc // ... repeat above code while (condition (when receive stop event, condition will became false value)) { [NSThread sleepForTimeInterval:10.0f]; [speechRecognizer stopContinuousRecognition]; }

after add this while loop, and it will recognize, then stop, this is expected (because i called stopContinuousRecognition).

My question is that why my recognition will stop automatically without calling stopContinuousRecognition ?
Rohit Mungi 49,131 Reputation points Microsoft Employee Moderator

2023-12-27T16:51:07.2733333+00:00

@玉洁朱 Referencing the same issue from SDK repo for future guidance.

https://github.com/Azure-Samples/cognitive-services-speech-sdk/issues/2192

AFAIK Unless stopContinuousRecognition is called your stream should not stop. I haven't used iOS SDK before to check if the same functionality works with the SDK, but the above referenced issue should provide an answer from SDK team.

1 answer

Your answer

Rohit Mungi 49,131 Reputation points Microsoft Employee Moderator

2023-12-27T10:22:44.84+00:00

@玉洁朱 Could you try to check if this sample works for you?
玉洁朱 0 Reputation points

2023-12-27T11:43:55.4+00:00

@Rohit Mungi Thanks for reply.

That sample works for me. But the difference with my project is that:

the sample code's speech source is come from a file named "whatstheweatherlike.wav". But my project's speech source is come from microphone.

Since my speech source is come from microphone, I need use [recorder] to record sound, and then offer the stream to Azure speech recognition, my code is like:

SPXPushAudioInputStream *stream = [[SPXPushAudioInputStream alloc] init]; AudioRecorder *recorder = [[AudioRecorder alloc]initWithPushStream:stream]; SPXAudioConfiguration *audioConfig = [[SPXAudioConfiguration alloc]initWithStreamInput:stream]; [recorder record]; [speechRecognizer addRecognizingEventHandler: xxxxxxx [speechRecognizer addRecognizedEventHandler: xxxxxxxx [speechRecognizer addSessionStoppedEventHandler: xxxxx [speechRecognizer startContinuousRecognition];

When I run my code, it will begin recognizing, but will soon stop immediately, and i will receive a session stop event.

I compare my code with your sample code, and found that sample code has a while loop, and write data to stream untile the source file read the end. So I also add a while loop :

objc // ... repeat above code while (condition (when receive stop event, condition will became false value)) { [NSThread sleepForTimeInterval:10.0f]; [speechRecognizer stopContinuousRecognition]; }

after add this while loop, and it will recognize, then stop, this is expected (because i called stopContinuousRecognition).

My question is that why my recognition will stop automatically without calling stopContinuousRecognition ?
Rohit Mungi 49,131 Reputation points Microsoft Employee Moderator

2023-12-27T16:51:07.2733333+00:00

@玉洁朱 Referencing the same issue from SDK repo for future guidance.

https://github.com/Azure-Samples/cognitive-services-speech-sdk/issues/2192

AFAIK Unless stopContinuousRecognition is called your stream should not stop. I haven't used iOS SDK before to check if the same functionality works with the SDK, but the above referenced issue should provide an answer from SDK team.

Answer 1

It seems you’re encountering an issue with the startContinuousRecognition function in the Microsoft Cognitive Services Speech SDK for iOS. Let’s troubleshoot this and find a solution:

Inconsistent Delays with stopContinuousRecognition:

The problem you’re facing is that when you call stopContinuousRecognition, there’s a 2-10 second delay that halts your entire program.
You’ve experimented with threading and async versions, but the inconsistency persists.
One possible cause is that the halting occurs when the stop function is called while a connection process is happening behind the scenes.
As a workaround, consider switching to using the synchronous versions of the start and stop functions from your background thread<sup>1</sup>.

Potential Solutions:

Restart SpeechRecognizer: When your app goes into the foreground (e.g., switching to another app), the Speech Recognizer may stop without any event. Try restarting the SpeechRecognizer when it comes back into the foreground<sup>2</sup>.
Check Threads: Ensure that you’re not running out of threads. Make sure you’re managing threads correctly and not blocking the main thread.
Verify Audio Stream Handling: Double-check how you’re handling the audio stream. Make sure you’re properly initializing, streaming, and closing the audio stream.
- Review SDK Documentation: Refer to the official SDK documentation for any specific guidelines or best practices related to continuous recognition.

Sample Code Modification:
- Consider modifying your code as follows:

      // Initialize SpeechRecognizer and other configurations
[speechRecognizer startContinuousRecognition];
[NSThread sleepForTimeInterval:3.0f]; // Add this code to delay stop
// Call [speechRecognizer stopContinuousRecognition] when needed
      
      ```

   - [If you don’t explicitly call `stopContinuousRecognition`, continuous recognition should not stop](https://github.com/Azure-Samples/cognitive-services-speech-sdk/issues/1506)<sup>[1](https://github.com/Azure-Samples/cognitive-services-speech-sdk/issues/1506)</sup>.

Remember to thoroughly test these changes and monitor the behavior. If the issue persists, consider reaching out to the Azure-Samples GitHub community for further assistance. 🌟🔧🎙️

Share via

【cognitive-services-speech-sdk】startContinuousRecognition will automatically stop

What I expected

1 answer

Your answer