Add files via upload

rvc_for_realtime.py

@@ -1,12 +1,12 @@
+import os, sys
 import faiss, torch, traceback, parselmouth, numpy as np, torchcrepe, torch.nn as nn, pyworld
-from fairseq import checkpoint_utils
+import fairseq
 from lib.infer_pack.models import (
     SynthesizerTrnMs256NSFsid,
     SynthesizerTrnMs256NSFsid_nono,
     SynthesizerTrnMs768NSFsid,
     SynthesizerTrnMs768NSFsid_nono,
 )
-import os, sys
 from time import time as ttime
 import torch.nn.functional as F
 import scipy.signal as signal
@@ -18,8 +18,14 @@ from multiprocessing import Manager as M
 
 mm = M()
 config = Config()
-
-
+if(config.dml==True):
+    def forward_dml(ctx, x, scale):
+        ctx.scale = scale
+        res = x.clone().detach()
+        return res
+    fairseq.modules.grad_multiply.GradMultiply.forward=forward_dml
+# config.device=torch.device("cpu")########强制cpu测试
+# config.is_half=False########强制cpu测试
 class RVC:
     def __init__(
         self, key, pth_path, index_path, index_rate, n_cpu, inp_q, opt_q, device
@@ -31,6 +37,7 @@ class RVC:
             global config
             self.inp_q = inp_q
             self.opt_q = opt_q
+            # device="cpu"########强制cpu测试
             self.device = device
             self.f0_up_key = key
             self.time_step = 160 / 16000 * 1000
@@ -46,7 +53,7 @@ class RVC:
                 self.big_npy = self.index.reconstruct_n(0, self.index.ntotal)
                 print("index search enabled")
             self.index_rate = index_rate
-            models, _, _ = checkpoint_utils.load_model_ensemble_and_task(
+            models, _, _ = fairseq.checkpoint_utils.load_model_ensemble_and_task(
                 ["hubert_base.pt"],
                 suffix="",
             )
@@ -80,6 +87,7 @@ class RVC:
             del self.net_g.enc_q
             print(self.net_g.load_state_dict(cpt["weight"], strict=False))
             self.net_g.eval().to(device)
+            # print(2333333333,device,config.device,self.device)#net_g是device，hubert是config.device
             if config.is_half:
                 self.net_g = self.net_g.half()
             else:
@@ -124,7 +132,7 @@ class RVC:
 
             pad_size = (p_len - len(f0) + 1) // 2
             if pad_size > 0 or p_len - len(f0) - pad_size > 0:
-                print(pad_size, p_len - len(f0) - pad_size)
+                # print(pad_size, p_len - len(f0) - pad_size)
                 f0 = np.pad(
                     f0, [[pad_size, p_len - len(f0) - pad_size]], mode="constant"
                 )
@@ -175,7 +183,10 @@ class RVC:
         return self.get_f0_post(f0bak)
 
     def get_f0_crepe(self, x, f0_up_key):
+        if self.device.type == "privateuseone":###不支持dml，cpu又太慢用不成，拿pm顶替
+            return self.get_f0(x, f0_up_key,1,"pm")
         audio = torch.tensor(np.copy(x))[None].float()
+        # print("using crepe,device:%s"%self.device)
         f0, pd = torchcrepe.predict(
             audio,
             self.sr,
@@ -184,6 +195,7 @@ class RVC:
             self.f0_max,
             "full",
             batch_size=512,
+            # device=self.device if self.device.type!="privateuseone" else "cpu",###crepe不用半精度全部是全精度所以不愁###cpu延迟高到没法用
             device=self.device,
             return_periodicity=True,
         )
@@ -197,10 +209,12 @@ class RVC:
     def get_f0_rmvpe(self, x, f0_up_key):
         if hasattr(self, "model_rmvpe") == False:
             from lib.rmvpe import RMVPE
-
             print("loading rmvpe model")
             self.model_rmvpe = RMVPE(
-                "rmvpe.pt", is_half=self.is_half, device=self.device
+                # "rmvpe.pt", is_half=self.is_half if self.device.type!="privateuseone" else False, device=self.device if self.device.type!="privateuseone"else "cpu"####dml时强制对rmvpe用cpu跑
+                #  "rmvpe.pt", is_half=False, device=self.device####dml配置
+                   # "rmvpe.pt", is_half=False, device="cpu"####锁定cpu配置
+                 "rmvpe.pt", is_half=self.is_half, device=self.device####正常逻辑
             )
             # self.model_rmvpe = RMVPE("aug2_58000_half.pt", is_half=self.is_half, device=self.device)
         f0 = self.model_rmvpe.infer_from_audio(x, thred=0.03)
@@ -279,6 +293,7 @@ class RVC:
         sid = torch.LongTensor([ii]).to(self.device)
         with torch.no_grad():
             if self.if_f0 == 1:
+                # print(12222222222,feats.device,p_len.device,cache_pitch.device,cache_pitchf.device,sid.device,rate2)
                 infered_audio = (
                     self.net_g.infer(
                         feats, p_len, cache_pitch, cache_pitchf, sid, rate2