Facing a problem with conversation transcriber with long audio file: canceled event after transcribing 1600 seconds.

Question

Hello, I am facing an issue with ConversationTranscriber method of speechsdk. It works for some audio files but there is one, of almost 2 hours in Japanese, where the event is canceled after transcribing ~1600 hours. When I checked the logs, I had this error :

[162383]: 1276687ms SPX_DBG_TRACE_VERBOSE: named_properties.h:479 ISpxPropertyBagImpl::SetStringValue: this=0x0x007ff9f80ea530; name='RESULT-ErrorDetails'; value='Due to service inactivity, the client buffer exceeded maximum size. Resetting the buffer.'[162383]: 1276692ms SPX_DBG_TRACE_VERBOSE: audio_stream_session.cpp:1397 [0xa23bda0]CSpxAudioStreamSession::WaitForRecognition_Complete: ...[162383]: 1276711ms SPX_DBG_TRACE_VERBOSE: audio_stream_session.cpp:1581 [0xa23bda0]CSpxAudioStreamSession::FireResultEvent[162383]: 1276724ms SPX_DBG_TRACE_VERBOSE: named_properties.h:479 ISpxNamedProperties::GetStringValue: this=0x0x007ff9f80ea530; name='RESULT-ErrorDetails'; value='Due to service inactivity, the client buffer exceeded maximum size. Resetting the buffer.'[382078]: 1276723ms SPX_DBG_TRACE_VERBOSE: audio_pump.cpp:446 [0xa06a880]CSpxAudioPump::PumpThread(): read frame duration: 9 ms => sending audio buffer size 6400[162383]: 1276733ms SPX_DBG_TRACE_VERBOSE: named_properties.h:479 ISpxPropertyBagImpl::SetStringValue: this=0x0x007ff9f80ea530; name='RESULT-ErrorDetails'; value='Due to service inactivity, the client buffer exceeded maximum size. Resetting the buffer. SessionId: b11e4d04e668435882fd7b33d438f897'[162383]: 1276741ms SPX_DBG_TRACE_VERBOSE: audio_stream_session.cpp:1472 [0xa23bda0]CSpxAudioStreamSession::FireSessionStoppedEvent: Firing SessionStopped event: SessionId: b11e4d04e668435882fd7b33d438f897[264420]: 1276741ms SPX_DBG_TRACE_SCOPE_ENTER: audio_stream_session.cpp:1724 DispatchEvent task started...[162383]: 1276748ms SPX_DBG_TRACE_SCOPE_ENTER: audio_stream_session.cpp:1560 EnsureFireResultEvent[162383]: 1276753ms SPX_DBG_TRACE_VERBOSE: audio_stream_session.cpp:1561 [0xa23bda0]CSpxAudioStreamSession::EnsureFireResultEvent[162383]: 1276758ms SPX_DBG_TRACE_SCOPE_EXIT: audio_stream_session.cpp:1560 EnsureFireResultEvent[162383]: 1276766ms SPX_DBG_TRACE_SCOPE_ENTER: audio_stream_session.cpp:1277 StopRecognizing[162383]: 1276773ms SPX_DBG_TRACE_VERBOSE: audio_stream_session.cpp:1278 [0xa23bda0]CSpxAudioStreamSession::StopRecognizing ...[162383]: 1276778ms SPX_DBG_TRACE_VERBOSE: audio_stream_session.cpp:3581 [0xa23bda0]CSpxAudioStreamSession::TryChangeState: recoKind/sessionState: 4/2 => 4/4[162383]: 1276784ms SPX_DBG_TRACE_VERBOSE: audio_stream_session.cpp:1356 [0xa23bda0]CSpxAudioStreamSession::StopRecognizing: We've been asked to stop whatever it is we're doing, while we're actively processing audio ...[162383]: 1276790ms SPX_DBG_TRACE_VERBOSE: audio_stream_session.cpp:1359 [0xa23bda0]CSpxAudioStreamSession::StopRecognizing: Now StoppingPump[0x9f9ec58] ...Here is the code used :

global AZURE_OUTPUTS
global SEG_COUNTER 
SEG_COUNTER = 1
AZURE_OUTPUTS = []

def conversation_transcriber_recognition_canceled_cb(evt: speechsdk.SessionEventArgs):
    print('Canceled event')

def conversation_transcriber_session_stopped_cb(evt: speechsdk.SessionEventArgs):
    print('SessionStopped event')

def processWords(paraDict):
    speaker = paraDict.get('SpeakerName', 'Unknown')
    text = paraDict.get('Text', '')

    return speaker, text

def conversation_transcriber_transcribed_cb(evt: speechsdk.SpeechRecognitionEventArgs):
    try:
       print('TRANSCRIBED:')
       if evt.result.reason == speechsdk.ResultReason.RecognizedSpeech:
           
           paraDict = dict()
           results = json.loads(evt.result.json)
           startValue = results['Offset']/1e7
           duration   = results['Duration']/1e7
           endValue   = duration + startValue
           displayText = results['DisplayText']
           speaker = results['SpeakerId']
           paraDict['SpeakerName'] = speaker
           paraDict['Text'] = displayText
           speaker, text = processWords(paraDict=paraDict)
           if speaker != "Unknown":
             speaker    = 'SPEAKER_'+str(int(speaker.split('-')[1])).zfill(2)
           
           print('\tText={}'.format(text))
           print('\tSpeaker ID={}'.format(speaker))
           print('\tOffset={}'.format(startValue))
           print('\tDuration={}'.format(duration))
           print('\tEndValue={}'.format(endValue))
           output = {'text':text, 'speaker':speaker, 'startValue':startValue, 'endValue':endValue}
           AZURE_OUTPUTS.append(output)
       elif evt.result.reason == speechsdk.ResultReason.NoMatch:
           print('\tNOMATCH: Speech could not be TRANSCRIBED: {}'.format(evt.result.no_match_details))
    except:
        print(traceback.format_exc())
        logging.error(traceback.format_exc())

def conversation_transcriber_session_started_cb(evt: speechsdk.SessionEventArgs):
    print('SessionStarted event')

def recognize_from_file(filename, language):

            import wave
            import contextlib
            with contextlib.closing(wave.open(filename,'r')) as f:
                frames = f.getnframes()
                rate = f.getframerate()
                duration = frames / float(rate)
                print(f'duration (seconds) : {duration}')
            # This example requires environment variables named "SPEECH_KEY" and "SPEECH_REGION"
            
            #### run source .env before running the script to get the environment variables
            
            speech_config = speechsdk.SpeechConfig(subscription=os.environ.get('SPEECH_KEY'), region=os.environ.get('SPEECH_REGION'))
            #speech_config = speechsdk.SpeechConfig(subscription='454e3547c097468f9be4771867b71af5', region='westeurope')
            speech_config.speech_recognition_language=language
            speech_config.enable_dictation()
            # Set conversation ending detection timeout (4 hours in seconds)     
            conversation_ending_detection_timeout = duration
                 
speech_config.set_service_property("speechsdk.PropertyId.Speech_SegmentationSilenceTimeoutMs", str(conversation_ending_detection_timeout * 1000), speechsdk.ServicePropertyChannel.UriQueryParameter)
           

            audio_config = speechsdk.audio.AudioConfig(filename=filename)
            conversation_transcriber = speechsdk.transcription.ConversationTranscriber(speech_config=speech_config, audio_config=audio_config)
            
            transcribing_stop = False

            def stop_cb(evt: speechsdk.SessionEventArgs):
                #"""callback that signals to stop continuous recognition upon receiving an event `evt`"""
                print('CLOSING on {}'.format(evt))
                nonlocal transcribing_stop
                transcribing_stop = True

            # Connect callbacks to the events fired by the conversation transcriber
            conversation_transcriber.transcribed.connect(conversation_transcriber_transcribed_cb)
            conversation_transcriber.session_started.connect(conversation_transcriber_session_started_cb)
            conversation_transcriber.session_stopped.connect(conversation_transcriber_session_stopped_cb)
            conversation_transcriber.canceled.connect(conversation_transcriber_recognition_canceled_cb)
            # stop transcribing on either session stopped or canceled events
            conversation_transcriber.session_stopped.connect(stop_cb)
            conversation_transcriber.canceled.connect(stop_cb)

            conversation_transcriber.start_transcribing_async()

            # Waits for completion.
            while not transcribing_stop:
                time.sleep(.5)

            conversation_transcriber.stop_transcribing_async()

filename = "xxxxxxxx.wav" #JP
if __name__ == "__main__":
    recognize_from_file(filename, 'ja-JP')