update

app.py

@@ -1,9 +1,8 @@
-import io
 import os
 import gradio as gr
 import librosa
 import numpy as np
-import soundfile
+import utils
 from inference.infer_tool import Svc
 import logging
 import webbrowser
@@ -40,6 +39,7 @@ def create_vc_fn(model, sid):
         if sampling_rate != 44100:
             audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=44100)
         out_audio, out_sr = model.infer(sid, vc_transform, audio, auto_predict_f0=auto_f0)
+        model.clear_empty()
         return "Success", (44100, out_audio.cpu().numpy())
     return vc_fn
 
@@ -50,11 +50,11 @@ if __name__ == '__main__':
     parser.add_argument("--share", action="store_true", default=False, help="share gradio app")
     parser.add_argument("--colab", action="store_true", default=False, help="share gradio app")
     args = parser.parse_args()
-
+    hubert_model = utils.get_hubert_model().to(args.device)
     models = []
     for f in os.listdir("models"):
         name = f
-        model = Svc(fr"models/{f}/{f}.pth", f"models/{f}/config.json", device=args.device)
+        model = Svc(fr"models/{f}/{f}.pth", f"models/{f}/config.json", device=args.device, hubert_model=hubert_model)
         cover = f"models/{f}/cover.png" if os.path.exists(f"models/{f}/cover.png") else None
         models.append((name, cover, create_vc_fn(model, name)))
     with gr.Blocks() as app:

inference/infer_tool.py

@@ -109,7 +109,7 @@ def split_list_by_n(list_collection, n, pre=0):
 
 
 class Svc(object):
-    def __init__(self, net_g_path, config_path,
+    def __init__(self, net_g_path, config_path, hubert_model,
                  device=None,
                  cluster_model_path="logs/44k/kmeans_10000.pt"):
         self.net_g_path = net_g_path
@@ -123,7 +123,7 @@ class Svc(object):
         self.hop_size = self.hps_ms.data.hop_length
         self.spk2id = self.hps_ms.spk
         # 加载hubert
-        self.hubert_model = utils.get_hubert_model().to(self.dev)
+        self.hubert_model = hubert_model
         self.load_model()
         if os.path.exists(cluster_model_path):
             self.cluster_model = cluster.get_cluster_model(cluster_model_path)

requirements.txt

@@ -1,6 +1,6 @@
 Flask
 Flask_Cors
-gradio
+gradio==3.18.0
 numpy
 playsound
 pydub

utils.py

@@ -244,7 +244,6 @@ def load_checkpoint(checkpoint_path, model, optimizer=None, skip_optimizer=False
         model.module.load_state_dict(new_state_dict)
     else:
         model.load_state_dict(new_state_dict)
-    print("load ")
     logger.info("Loaded checkpoint '{}' (iteration {})".format(
         checkpoint_path, iteration))
     return model, optimizer, learning_rate, iteration