Update app.py

app.py

@@ -43,9 +43,6 @@ model.load_checkpoint(config, checkpoint_path=checkpoint_path, vocab_path=vocab_
 
 print("Modelo cargado en CPU")
 
-def split_text(text):
-    return re.split(r'(?<=[.!?])\s+', text)
-
 def predict(prompt, language, reference_audio):
     try:
         if len(prompt) < 2 or len(prompt) > 600:
@@ -53,22 +50,15 @@ def predict(prompt, language, reference_audio):
 
         sentences = split_text(prompt)
 
-        # Usar los parámetros del config.json
-        temperature = config.model_args.get("temperature", 0.85)
-        repetition_penalty = config.model_args.get("repetition_penalty", 2.0)
-        length_penalty = config.model_args.get("length_penalty", 1.0)
-        top_k = config.model_args.get("top_k", 50)
-        top_p = config.model_args.get("top_p", 0.85)
-        
-        gpt_cond_len = config.model_args.get("gpt_cond_len", 12)
-        gpt_cond_chunk_len = config.model_args.get("gpt_cond_chunk_len", 4)
-        max_ref_len = config.model_args.get("max_ref_len", 10)
+        # Obtener los parámetros de la configuración JSON
+        temperature = config.get("temperature", 0.85)
+        length_penalty = config.get("length_penalty", 1.0)
+        repetition_penalty = config.get("repetition_penalty", 2.0)
+        top_k = config.get("top_k", 50)
+        top_p = config.get("top_p", 0.85)
 
         gpt_cond_latent, speaker_embedding = model.get_conditioning_latents(
-            audio_path=reference_audio,
-            gpt_cond_len=gpt_cond_len,
-            gpt_cond_chunk_len=gpt_cond_chunk_len,
-            max_ref_len=max_ref_len
+            audio_path=reference_audio
         )
 
         start_time = time.time()