Update app.py

app.py

@@ -13,16 +13,23 @@ MODEL_REPO = "grandhigh/Chatterbox-TTS-Indonesian"
 CHECKPOINT_FILENAME = "t3_cfg.safetensors"
 DEVICE = "cpu"
 
-print("Loading model...")
-model = ChatterboxTTS.from_pretrained(device=DEVICE)
+_model = None
 
-checkpoint_path = hf_hub_download(repo_id=MODEL_REPO, filename=CHECKPOINT_FILENAME)
-t3_state = load_file(checkpoint_path, device="cpu")
-model.t3.load_state_dict(t3_state)
+def get_model():
+    global _model
+    if _model is None:
+        print("Loading model on first request...")
+        m = ChatterboxTTS.from_pretrained(device=DEVICE)
 
-model = model.to(DEVICE)
-model.eval()
-print("Model loaded.")
+        checkpoint_path = hf_hub_download(repo_id=MODEL_REPO, filename=CHECKPOINT_FILENAME)
+        t3_state = load_file(checkpoint_path, device="cpu")
+        m.t3.load_state_dict(t3_state)
+
+        m = m.to(DEVICE)
+        m.eval()
+        _model = m
+        print("Model loaded.")
+    return _model
 
 def _download_audio_from_url(url: str) -> str:
     r = requests.get(url, timeout=60)
@@ -45,6 +52,8 @@ def clone_voice(text: str, audio_file, audio_url: str):
     if prompt_path is None:
         raise gr.Error("Upload WAV atau isi audio_url.")
 
+    model = get_model()
+
     with torch.no_grad():
         wav = model.generate(text.strip(), audio_prompt_path=prompt_path)
 
@@ -71,5 +80,6 @@ with gr.Blocks(title="Chatterbox Indonesian Voice Cloning API") as demo:
     )
 
 if __name__ == "__main__":
-    port = int(os.getenv("PORT", os.getenv("GRADIO_SERVER_PORT", 7860)))
-    demo.launch(server_name="0.0.0.0", server_port=port)
+    port = int(os.getenv("PORT", "7860"))
+    demo.queue(default_concurrency_limit=1)
+    demo.launch(server_name="0.0.0.0", server_port=port, show_error=True)