Refactor model loading in app.py and update TTS version in requirements.txt

app.py

@@ -28,23 +28,20 @@ if not speaker_audio_path.exists():
 
 config_path = str(config_path)
 vocab_path = str(vocab_path)
-model_path = str(model_path.parent)
+model_path = str(model_path)
 speaker_audio_path = str(speaker_audio_path)
 
-model = None
+config = XttsConfig()
+config.load_json(config_path)
+
+print("Loading model...")
+device = "cuda" if torch.cuda.is_available() else "cpu"
+model = Xtts.init_from_config(config)
+model.load_checkpoint(config, checkpoint_path=model_path, use_deepspeed=True, vocab_path=vocab_path, eval=True)
+model.to(device)
 
 @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])
 

requirements.txt

@@ -1,4 +1,4 @@
-git+https://github.com/coqui-ai/TTS
+TTS @ git+https://github.com/coqui-ai/TTS@v0.21.1
 transformers
 deepspeed
 triton