[Python] Python streamlit realtime speech-to-text with azure SDK

Hello I'm trying to create a real-time speech to text using streamlit and azure speech SDK. I can easilly transcribe audio/video files with no issues, but I want to integrate realtime transcription (from browser)

In the code I've tryed when I speak in the micropghone no sentence is transcribed

I've also tryed to reuse the function I've created that uses files, pass the AudioStream and turn it async but didn't work it either also the guided path: https://microsoft.github.io/TechExc...cs/04_implement_audio_transcription/0402.html Only works on local machine because it uses host's microphone

I've tryed with the code posted blow, search on google and ask AI. I want the user to be able to start live speech-to-text with live transcription in the chat (speaker recognition must stay)

Code:

webrtc_ctx = webrtc_streamer(key="speech-to-text", mode=WebRtcMode.SENDONLY,
media_stream_constraints={"video": False, "audio": True})

stream= PushAudioInputStream()

audio_input = speechsdk.AudioConfig(stream=stream)
speech_config = speechsdk.SpeechConfig(env["SPEECH_KEY"], env["SPEECH_REGION"])
#other settings
conversation_transcriber = speechsdk.transcription.ConversationTranscriber(speech_config, audio_input)

stream_placeholder = None; done = False; r = []

def stop_cb(evt: ConversationTranscriptionEventArgs):
conversation_transcriber.stop_transcribing_async()
logger.info(evt)
if evt.result.reason == speechsdk.ResultReason.Canceled:
canc = evt.result.cancellation_details
logger.warning(canc)
if canc.reason == speechsdk.CancellationReason.Error:
logger.error("Error details: {}".format(canc.error_details))
raise Exception("Error details: {}".format(canc.error_details))

global done
done = True

def addsentence(evt: ConversationTranscriptionEventArgs):
global r
if evt.result.speaker_id == "Unknown":
logger.debug("Unknown speaker: " + str(evt))
return
logger.debug(f"Detected **{evt.result.speaker_id}**: {evt.result.text}")
r.append(f"**{evt.result.speaker_id}**: {evt.result.text}")

conversation_transcriber.transcribed.connect(addsentence)
conversation_transcriber.session_stopped.connect(stop_cb)
conversation_transcriber.canceled.connect(stop_cb)

while webrtc_ctx.state.playing:
if not st.session_state["recording"]:
with st.chat_message("assistant"):
with st.spinner("Starting transcription..."):
stream_placeholder = st.expander("Transcription", icon="").empty()

conversation_transcriber.start_transcribing_async()
logger.info("Transcribing started!")
st.session_state["recording"] = True

logger.debug("Receiving frame from browser...")
try:
audio_frames = webrtc_ctx.audio_receiver.get_frames(timeout=1)
except queue.Empty:
time.sleep(0.1)
logger.debug("No frame arrived.")
continue

stream_placeholder.markdown("## Transcription:\n\n" + "\\\n".join(r))

for audio_frame in audio_frames:
stream.write(audio_frame.to_ndarray().tobytes())

Continue reading...