How to increase the time for which the Microsoft Speech Service SDK listens in a single go?

Abdullah Nadeem 0

I am using MS speech service sdk for speech to text conversion. When I speak, my speech is converted to text after 60 seconds even if I haven't stopped speaking. It basically considers it one chunk and starts processing it. What can I do to increase this timeout?

I want the service to keep listening even if I speak for 80 seconds and convert the speech to text only after I have finished speaking.

santoshkc 4,425 Microsoft Vendor

Hi @Abdullah Nadeem,

Thank you for reaching out to Microsoft Q&A forum!

I tried to repro your issue, but the approach suggested (for 80 seconds) might not be working as expected. But you can use the below code for continuous recognition, it will transcribe your audio until it's stopped.

import azure.cognitiveservices.speech as speechsdk
def continuous_recognition_from_microphone():
    # This example requires environment variables named "SPEECH_KEY" and "SPEECH_REGION"
    speech_config = speechsdk.SpeechConfig(subscription='SPEECH_KEY', region='SPEECH_REGION')
    speech_config.speech_recognition_language = "en-US"
    audio_config = speechsdk.audio.AudioConfig(use_default_microphone=True)
    speech_recognizer = speechsdk.SpeechRecognizer(speech_config=speech_config, audio_config=audio_config)
    print("Speak into your microphone.")
    # Event handler for recognizing speech
    def recognized_callback(evt):
        result = evt.result
        if result.reason == speechsdk.ResultReason.RecognizedSpeech:
            print("Recognized: {}".format(result.text))
        elif result.reason == speechsdk.ResultReason.NoMatch:
            print("No speech could be recognized: {}".format(result.no_match_details))
        elif result.reason == speechsdk.ResultReason.Canceled:
            cancellation_details = result.cancellation_details
            print("Speech Recognition canceled: {}".format(cancellation_details.reason))
            if cancellation_details.reason == speechsdk.CancellationReason.Error:
                print("Error details: {}".format(cancellation_details.error_details))
                print("Did you set the speech resource key and region values?")
    # Connect the event handler
    speech_recognizer.recognized.connect(recognized_callback)
    # Start continuous recognition
    speech_recognizer.start_continuous_recognition()
    input("Press Enter to stop recognition...\n")
    # Stop continuous recognition
    speech_recognizer.stop_continuous_recognition()
continuous_recognition_from_microphone()

I hope you understand! Thank you.

Abdullah Nadeem 0 Reputation points

2024-03-20T01:58:30.94+00:00

Thank you for replying @santoshkc . I have implemented it the same way in my React app. But the thing is, this does not do I want it to do. This does not stop continuous recognition until you press Enter. But if you keep speaking for over 60 seconds, it will process the first 60 seconds of audio as one chunk, and then the next 60 seconds of audio as the next chunk. I want it to process the whole 80 or 120 seconds of audio as one chunk.

In short, I want it to keep recognizing until I stop speaking, and only then give me the final recognized text.
santoshkc 4,425 Reputation points Microsoft Vendor

2024-03-20T12:25:24.4866667+00:00

Hi @Abdullah Nadeem,

Thank you for your follow-up question.

Unfortunately, the Speech Service SDK does not support continuous recognition until the user stops speaking. However, you can use the recognize_once_async() method with a file input stream instead of a microphone input stream to recognize speech from a file. This method allows you to process large audio files as a single chunk, which should meet your requirement of processing the whole 80 or 120 seconds of audio as one chunk. See Recognize speech from a file.

I hope this helps! Thank you.
Abdul Moiz Tauqir 0 Reputation points

2024-04-24T07:11:44.39+00:00
Hi @santoshkc ,
We are still encountering the same issue with Python Speech SDK. We have attempted to configure the response times using the start_continuous_recognition_async() function with the following time parameters:

InitialSilenceTimeoutMs

Speech_SegmentationSilenceTimeoutMs

SpeechServiceConnection_EndSilenceTimeoutMs

Despite setting these parameters to 120 seconds, we have not been able to achieve the desired outcome. We continue to receive responses for recognized state for a maximum of 60 seconds of audio to speech.
Could you please guide us to the parameter that needs to be adjusted to effectively increase the response duration beyond 60 seconds? Alternatively, could you direct us to the documentation where we can find this detail?
I am assuming there is no limitation on the service side to restrict a recognized state response to a maximum of 60 seconds.
Thank you.