Update app.py

app.py

@@ -19,17 +19,21 @@ def generate_text(audio):
     if audio is None:
         return "No audio received"
     
-    sr, data = audio  # Unpack the tuple
+    sr, data = audio  # Unpack the tuple (sample rate, numpy array)
     print(f"Original sample rate: {sr}, dtype: {data.dtype}")
 
-    # Convert to float32 and ensure 16kHz
+    # Convert to float32
     data = data.astype(np.float32)
-    if sr != 16000:
-        data = librosa.resample(data, orig_sr=sr, target_sr=16000)
-        sr = 16000
+
+    # Resample to 16kHz if necessary
+    target_sr = 16000
+    if sr != target_sr:
+        data = librosa.resample(data, orig_sr=sr, target_sr=target_sr)
+        sr = target_sr
 
     print(f"Processed sample rate: {sr}, dtype: {data.dtype}")
 
+    # Tokenize and run inference
     inputs = asr_processor(data, sampling_rate=sr, return_tensors="pt", padding=True)
     
     with torch.no_grad():
@@ -39,12 +43,10 @@ def generate_text(audio):
     # Decode the prediction
     return asr_processor.decode(pred_ids)
 
-
-# Ensure Gradio records at 16kHz and float32
+# Ensure we get a NumPy array from Gradio
 input_audio = gr.Audio(
     sources=["microphone"],
-    type="numpy",  # Ensures we receive (sr, np.ndarray)
-    sample_rate=16000,  # Force 16kHz recording
+    type="numpy",  # Ensures function gets (sr, np.ndarray)
 )
 
 demo = gr.Interface(