Update app.py

app.py

@@ -126,6 +126,7 @@ pipe_asr = pipeline(
     return_timestamps=True
 )
 
+# Function to handle audio input, transcribe, fetch from Neo4j, and generate audio response
 # Function to handle audio input, transcribe, fetch from Neo4j, and generate audio response
 def transcribe_and_respond(audio):
     if audio is None:
@@ -135,10 +136,12 @@ def transcribe_and_respond(audio):
     sr, y = audio
     y = np.array(y).astype(np.float32)
 
-    # Normalize the audio array
-    max_abs_y = np.max(np.abs(y))
-    if max_abs_y > 0:
-        y = y / max_abs_y
+    # Resample to 16kHz if needed
+    target_sr = 16000
+    if sr != target_sr:
+        logging.debug(f"Resampling audio from {sr} Hz to {target_sr} Hz.")
+        y = torchaudio.functional.resample(torch.tensor(y), orig_freq=sr, new_freq=target_sr).numpy()
+        sr = target_sr
 
     # Prepare input_features for Whisper model
     input_features = processor(y, sampling_rate=sr, return_tensors="pt").input_features