Update app.py

app.py

@@ -35,28 +35,33 @@ def transcribe_audio(audio_path):
     # Load audio
     speech, sr = librosa.load(audio_path, sr=16000)
 
+    # Convert to batch of shape (1, seq_len)
     inputs = processor(
         audios=speech,
         sampling_rate=16000,
+        # specify target language here
+        language="yo",          # Yoruba ISO-639-1 code
         return_tensors="pt"
-    ).to(DEVICE)
+    )
+
+    # Move input_features to device
+    input_features = inputs["input_features"].to(DEVICE)
 
     with torch.no_grad():
-        predicted_ids = asr_model.generate(
-            inputs["input_features"],
-            task="transcribe",   # force transcription
-            language="yo",       # Yoruba ISO-639-1 code
-            max_new_tokens=300
-        )
+        predicted_ids = asr_model.generate(input_features, max_new_tokens=300)
 
     transcription = processor.batch_decode(
         predicted_ids,
         skip_special_tokens=True
     )[0]
 
+    if not transcription.strip():
+        return "Could not transcribe audio. Please try again in clear Yoruba."
+
     return transcription.strip()
 
 
+
 demo = gr.Interface(
     fn=transcribe_audio,
     inputs=gr.Audio(type="filepath", label="Upload Speech"),