Update app.py

app.py

@@ -13,7 +13,6 @@ from inference import OmniInference
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 omni_client = OmniInference('./checkpoint', device)
-omni_client.warm_up()
 
 
 OUT_CHUNK = 4096
@@ -21,6 +20,10 @@ OUT_RATE = 24000
 OUT_CHANNELS = 1
 
 
+@spaces.GPU
+def warmup():
+    omni_client.warm_up()
+
 @spaces.GPU
 def process_audio(audio):
     filepath = audio