remove unused variables and functions

ddPn08 · ddPn08 · commit a488c7e4eb27 · 2023-05-28T11:26:54.000+09:00
diff --git a/lib/rvc/pipeline.py b/lib/rvc/pipeline.py
@@ -11,8 +11,6 @@
 
 # from faiss.swigfaiss_avx2 import IndexIVFFlat # cause crash on windows' faiss-cpu installed from pip
 from fairseq.models.hubert import HubertModel
-from transformers import HubertModel as TrHubertModel
-from transformers import Wav2Vec2FeatureExtractor
 
 from .models import SynthesizerTrnMs256NSFSid
 
@@ -61,7 +59,6 @@ def get_f0(
         f0_method: str,
         inp_f0: np.ndarray = None,
     ):
-        time_step = self.window / self.sr * 1000
         f0_min = 50
         f0_max = 1100
         f0_mel_min = 1127 * np.log(1 + f0_min / 700)
@@ -114,7 +111,7 @@ def get_f0(
 
     def _convert(
         self,
-        model: Union[HubertModel, Tuple[Wav2Vec2FeatureExtractor, TrHubertModel]],
+        model: HubertModel,
         embedding_output_layer: int,
         net_g: SynthesizerTrnMs256NSFSid,
         sid: int,
@@ -231,7 +228,7 @@ def _convert(
 
     def __call__(
         self,
-        model: Union[HubertModel, Tuple[Wav2Vec2FeatureExtractor, TrHubertModel]],
+        model: HubertModel,
         embedding_output_layer: int,
         net_g: SynthesizerTrnMs256NSFSid,
         sid: int,
diff --git a/lib/rvc/preprocessing/extract_f0.py b/lib/rvc/preprocessing/extract_f0.py
@@ -19,7 +19,6 @@ def compute_f0(
     f0_min: float,
 ):
     x = load_audio(path, fs)
-    p_len = x.shape[0] // hop
     if f0_method == "harvest":
         f0, t = pyworld.harvest(
             x.astype(np.double),
diff --git a/lib/rvc/preprocessing/extract_feature.py b/lib/rvc/preprocessing/extract_feature.py
@@ -10,8 +10,6 @@
 import torch.nn.functional as F
 from fairseq import checkpoint_utils
 from tqdm import tqdm
-from transformers import HubertModel as TrHubertModel
-from transformers import Wav2Vec2FeatureExtractor
 
 
 def load_embedder(embedder_path: str, device):
@@ -34,42 +32,6 @@ def load_embedder(embedder_path: str, device):
     return embedder_model, cfg
 
 
-def load_transformers_hubert(repo_name: str, device):
-    try:
-        feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(repo_name)
-        embedder_model = TrHubertModel.from_pretrained(repo_name).to(device)
-        if device != "cpu":
-            embedder_model = embedder_model.half()
-        else:
-            embedder_model = embedder_model.float()
-        embedder_model.eval()
-    except Exception as e:
-        print(f"Error: {e} {repo_name}")
-        traceback.print_exc()
-
-    return (feature_extractor, embedder_model), None
-
-
-def load_transformers_hubert_local(embedder_path: str, device):
-    try:
-        feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(
-            embedder_path, local_files_only=True
-        )
-        embedder_model = TrHubertModel.from_pretrained(
-            embedder_path, local_files_only=True
-        ).to(device)
-        if device != "cpu":
-            embedder_model = embedder_model.half()
-        else:
-            embedder_model = embedder_model.float()
-        embedder_model.eval()
-    except Exception as e:
-        print(f"Error: {e} {embedder_path}")
-        traceback.print_exc()
-
-    return (feature_extractor, embedder_model), None
-
-
 # wave must be 16k, hop_size=320
 def readwave(wav_path, normalize=False):
     wav, sr = sf.read(wav_path)
@@ -104,12 +66,7 @@ def processor(
     if embedder_load_from == "local" and not os.path.exists(embedder_path):
         return f"Embedder not found: {embedder_path}"
 
-    if embedder_load_from == "hf":
-        model, cfg = load_transformers_hubert(embedder_path, device)
-    elif embedder_load_from == "tr-local":
-        model, cfg = load_transformers_hubert_local(embedder_path, device)
-    else:
-        model, cfg = load_embedder(embedder_path, device)
+    model, cfg = load_embedder(embedder_path, device)
 
     for file in tqdm(todo, position=1 + process_id):
         try:
diff --git a/modules/models.py b/modules/models.py
@@ -6,8 +6,6 @@
 from fairseq import checkpoint_utils
 from fairseq.models.hubert.hubert import HubertModel
 from pydub import AudioSegment
-from transformers import HubertModel as TrHubertModel
-from transformers import Wav2Vec2FeatureExtractor
 
 from lib.rvc.models import SynthesizerTrnMs256NSFSid, SynthesizerTrnMs256NSFSidNono
 from lib.rvc.pipeline import VocalConvertPipeline
@@ -20,16 +18,12 @@
 
 
 EMBEDDINGS_LIST = {
-    # "hubert_base": ("hubert_base.pt", "hubert_base", "local"),
     "hubert-base-japanese": (
         "rinna_hubert_base_jp.pt",
         "hubert-base-japanese",
         "local",
     ),
     "contentvec": ("checkpoint_best_legacy_500.pt", "contentvec", "local"),
-    # "distilhubert": ("ntu-spml/distilhubert", "distilhubert", "hf"),
-    # "distilhubert-ja": ("TylorShine/distilhubert-ft-japanese-50k", "distilhubert-ja", "hf"),
-    # "distilhubert-ja_dev": ("models/pretrained/feature_extractors/distilhubert-ja-en", "distilhubert-ja_dev", "tr-local"),
 }
 
 
@@ -129,10 +123,13 @@ def single(
             )
             if embedder_model_name.endswith("768"):
                 embedder_model_name = embedder_model_name[:-3]
+
         if embedder_model_name == "hubert_base":
             embedder_model_name = "contentvec"
+
         if not embedder_model_name in EMBEDDINGS_LIST.keys():
             raise Exception(f"Not supported embedder: {embedder_model_name}")
+
         if (
             embedder_model == None
             or loaded_embedder_model != EMBEDDINGS_LIST[embedder_model_name][1]
@@ -141,12 +138,7 @@ def single(
             embedder_filename, embedder_name, load_from = get_embedder(
                 embedder_model_name
             )
-            if load_from == "hf":
-                load_transformers_hubert(embedder_filename, embedder_name)
-            elif load_from == "tr-local":
-                load_transformers_hubert_local(embedder_filename, embedder_name)
-            else:
-                load_embedder(embedder_filename, embedder_name)
+            load_embedder(embedder_filename, embedder_name)
 
         if embedding_output_layer == "auto":
             embedding_output_layer = (
@@ -255,41 +247,6 @@ def load_embedder(emb_file: str, emb_name: str):
     loaded_embedder_model = emb_name
 
 
-def load_transformers_hubert(repo_name: str, emb_name: str):
-    global embedder_model, loaded_embedder_model
-    feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(repo_name)
-    embedder = TrHubertModel.from_pretrained(repo_name).to(device)
-
-    if is_half:
-        embedder = embedder.half()
-    else:
-        embedder = embedder.float()
-    embedder.eval()
-
-    embedder_model = (feature_extractor, embedder)
-
-    loaded_embedder_model = emb_name
-
-
-def load_transformers_hubert_local(emb_file: str, emb_name: str):
-    global embedder_model, loaded_embedder_model
-    emb_file = os.path.join(ROOT_DIR, emb_file)
-    feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(
-        emb_file, local_files_only=True
-    )
-    embedder = TrHubertModel.from_pretrained(emb_file, local_files_only=True).to(device)
-
-    if is_half:
-        embedder = embedder.half()
-    else:
-        embedder = embedder.float()
-    embedder.eval()
-
-    embedder_model = (feature_extractor, embedder)
-
-    loaded_embedder_model = emb_name
-
-
 def get_vc_model(model_name: str):
     model_path = os.path.join(MODELS_DIR, "checkpoints", model_name)
     weight = torch.load(model_path, map_location="cpu")
