Update core.py

rvc_inferpy/core.py

@@ -20,10 +20,11 @@
 from multiprocessing import cpu_count
 from urllib.parse import urlparse
 from io import BytesIO
-from fairseq.models.hubert import HubertModel
 import torch
 import numpy as np
 import soundfile as sf
+import torchaudio
+from torchaudio.pipelines import HUBERT_BASE
 
 # Third-party package imports from rvc_inferpy
 from rvc_inferpy.split_audio import (
@@ -163,25 +164,23 @@ def note_to_hz(note_name: str) -> float | None:
 
 
 
+
 def load_hubert(config, hubert_path: str = None):
-
     """
-    Load and return the Hubert model using the HubertModel.from_pretrained API.
-    If hubert_path is not provided or does not exist, fallback to a default model.
+    Load and return the HuBERT model using torchaudio.
+    If the specified hubert_path does not exist, use the pre-trained torchaudio model.
     """
     if hubert_path is None or not os.path.exists(hubert_path):
-        # You can implement download logic here if needed.
-        hubert_path = "hubert_base.pt"  # Default model file name
-
-    hubert_model = HubertModel.from_pretrained(
-        model_name_or_path=".",            # Directory containing the model
-        checkpoint_file=hubert_path,         # The checkpoint file to load
-        data_name_or_path="."                # Dummy or actual data directory as required
-    )
-
+        print("Using torchaudio's pre-trained HuBERT model.")
+        hubert_model = HUBERT_BASE.get_model()
+    else:
+        print(f"Loading HuBERT model from {hubert_path}")
+        hubert_model = torch.jit.load(hubert_path)
+
     hubert_model = hubert_model.to(config.device)
     hubert_model = hubert_model.half() if config.is_half else hubert_model.float()
     hubert_model.eval()
+
     return hubert_model