Update main.py

main.py

@@ -168,7 +168,11 @@ async def transcribe_and_match(
  contents = await file.read()
  audio = AudioSegment.from_file(BytesIO(contents))
 
- # Step 2: Export to WAV format and load with torchaudio
+ # Step 2: Ensure audio is mono
+ if audio.channels > 1:
+ audio = audio.set_channels(1) # Convert to mono
+
+ # Step 3: Export to WAV format and load with torchaudio
  wav_buffer = BytesIO()
  audio.export(wav_buffer, format="wav")
  wav_buffer.seek(0)
@@ -179,14 +183,14 @@ async def transcribe_and_match(
  # Convert waveform to float32
  samples = waveform.numpy().astype(np.float32)
 
- # Step 3: Use the speech-to-text model
+ # Step 4: Use the speech-to-text model
  transcription_result = nlp_speech_to_text(samples)
  transcription_text = transcription_result['text']
 
- # Step 4: Parse the field_data (which contains field names/IDs)
+ # Step 5: Parse the field_data (which contains field names/IDs)
  fields = json.loads(field_data)
 
- # Step 5: Find the matching field for the transcription
+ # Step 6: Find the matching field for the transcription
  field_matches = {}
  for field in fields:
  field_label = field.get("field_label", "").lower()
@@ -196,7 +200,7 @@ async def transcribe_and_match(
  if field_label in transcription_text.lower():
  field_matches[field_id] = transcription_text
 
- # Step 6: Return transcription + matched fields
+ # Step 7: Return transcription + matched fields
  return {
  "transcription": transcription_text,
  "matched_fields": field_matches