Update main.py

main.py

@@ -8,6 +8,7 @@ from starlette.middleware import Middleware
 from starlette.middleware.cors import CORSMiddleware
 from pdf2image import convert_from_bytes
 from pydub import AudioSegment
+import numpy as np
 
 app = FastAPI()
 
@@ -165,22 +166,25 @@ async def transcribe_and_match(
         contents = await file.read()
         audio = AudioSegment.from_file(BytesIO(contents))
 
-        # Optionally convert to wav if needed
-        wav_io = BytesIO()
-        audio.export(wav_io, format="wav")
-        wav_io.seek(0)
+        # Convert AudioSegment to a NumPy array
+        # First, export to raw audio format and then load into NumPy
+        raw_audio = BytesIO()
+        audio.export(raw_audio, format="wav")
+        raw_audio.seek(0)
 
-        # Transcribe the WAV audio file
-        transcription_result = nlp_speech_to_text(wav_io)
+        # Convert audio to samples as NumPy array
+        samples = np.array(audio.get_array_of_samples())
+
+        # Step 2: Use the speech-to-text model (expecting NumPy array)
+        transcription_result = nlp_speech_to_text(samples)
         transcription_text = transcription_result['text']
 
-        # Step 2: Parse the field_data (which contains field names/IDs)
+        # Step 3: Parse the field_data (which contains field names/IDs)
         import json
         fields = json.loads(field_data)
 
-        # Step 3: Find the matching field for the transcription
+        # Step 4: Find the matching field for the transcription
         field_matches = {}
-
         for field in fields:
             field_label = field.get("field_label", "").lower()
             field_id = field.get("field_id", "")
@@ -189,7 +193,7 @@ async def transcribe_and_match(
             if field_label in transcription_text.lower():
                 field_matches[field_id] = transcription_text
 
-        # Step 4: Return transcription + matched fields
+        # Step 5: Return transcription + matched fields
         return {
             "transcription": transcription_text,
             "matched_fields": field_matches