Lazy load the model in infer_EGTTS to improve performance

app.py

@@ -31,16 +31,20 @@ vocab_path = str(vocab_path)
 model_path = str(model_path.parent)
 speaker_audio_path = str(speaker_audio_path)
 
-print("Loading model...")
-device = "cuda" if torch.cuda.is_available() else "cpu"
-config = XttsConfig()
-config.load_json(config_path)
-model = Xtts.init_from_config(config)
-model.load_checkpoint(config, checkpoint_dir=model_path, use_deepspeed=True, vocab_path=vocab_path)
-model.to(device)
+model = None
 
 @spaces.GPU
 def infer_EGTTS(text: str, speaker_audio_path: str, temperature: float = 0.75):
+    global model
+    if model is None:
+        print("Loading model...")
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+        config = XttsConfig()
+        config.load_json(config_path)
+        model = Xtts.init_from_config(config)
+        model.load_checkpoint(config, checkpoint_dir=model_path, use_deepspeed=True, vocab_path=vocab_path)
+        model.to(device)
+
     print("Computing speaker latents...")
     gpt_cond_latent, speaker_embedding = model.get_conditioning_latents(audio_path=[speaker_audio_path])