Update unified_tts.py

unified_tts.py

@@ -65,7 +65,7 @@ from transformers import AutoTokenizer
 from vllm import LLM, SamplingParams
 from modelscope import snapshot_download
 
-snapshot_download('iic/CosyVoice2-0.5B', local_dir='pretrained_models/CosyVoice2-0.5B')
+cosyvoice_cache_dir = snapshot_download('iic/CosyVoice2-0.5B') 
 
 
 # Add CosyVoice paths
@@ -212,7 +212,7 @@ class UnifiedTTS:
         # CosyVoice2 converts discrete speech tokens to continuous audio waveforms
         # It supports voice cloning using prompt audio features for consistent voice characteristics
         logger.info(f"Loading CosyVoice2 model from {self.cosyvoice_model_dir}")
-        self.cosyvoice = CosyVoice2(self.cosyvoice_model_dir, fp16=self.fp16)
+        self.cosyvoice = CosyVoice2(cosyvoice_cache_dir, fp16=self.fp16, device="cuda" if torch.cuda.is_available() else "cpu")
         self.sample_rate = self.cosyvoice.sample_rate
         
         # Preload prompt audio features for voice cloning