fix: MacOS 纯 CPU 推理时 Segmentation fault: 11

see: facebookresearch/faiss#2317 facebookresearch#2410

Retrieval_based_Voice_Conversion_WebUI.ipynb

@@ -129,10 +129,10 @@
         "#@title 从谷歌云盘加载打包好的数据集到/content/dataset\n",
         "\n",
         "#@markdown 数据集位置\n",
-        "DATASET = \"/content/drive/MyDrive/dataset/lulucall_48k.zip\"  #@param {type:\"string\"}\n",
+        "DATASET = \"/content/drive/MyDrive/dataset/lulu20230327_32k.zip\"  #@param {type:\"string\"}\n",
         "\n",
         "!mkdir -p /content/dataset\n",
-        "!unzip -d /content/dataset {DATASET}"
+        "!unzip -d /content/dataset -B {DATASET}"
       ],
       "metadata": {
         "id": "Mwk7Q0Loqzjx"
@@ -140,13 +140,26 @@
       "execution_count": null,
       "outputs": []
     },
+    {
+      "cell_type": "code",
+      "source": [
+        "#@title 重命名数据集中的重名文件\n",
+        "!ls -a /content/dataset/\n",
+        "!rename 's/(\\w+)\\.(\\w+)~(\\d*)/$1_$3.$2/' /content/dataset/*.*~*"
+      ],
+      "metadata": {
+        "id": "PDlFxWHWEynD"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
     {
       "cell_type": "code",
       "source": [
         "#@title 启动web\n",
         "%cd /content/Retrieval-based-Voice-Conversion-WebUI\n",
-        "%load_ext tensorboard\n",
+        "# %load_ext tensorboard\n",
-        "%tensorboard --logdir /content/Retrieval-based-Voice-Conversion-WebUI/logs\n",
+        "# %tensorboard --logdir /content/Retrieval-based-Voice-Conversion-WebUI/logs\n",
         "!python3 infer-web.py --colab --pycmd python3"
       ],
       "metadata": {
@@ -164,7 +177,7 @@
         "#@markdown 模型名\n",
         "MODELNAME = \"lulu\"  #@param {type:\"string\"}\n",
         "#@markdown 模型epoch\n",
-        "MODELEPOCH = 7500  #@param {type:\"integer\"}\n",
+        "MODELEPOCH = 9600  #@param {type:\"integer\"}\n",
         "\n",
         "!cp /content/Retrieval-based-Voice-Conversion-WebUI/logs/{MODELNAME}/G_{MODELEPOCH}.pth /content/drive/MyDrive/{MODELNAME}_D_{MODELEPOCH}.pth\n",
         "!cp /content/Retrieval-based-Voice-Conversion-WebUI/logs/{MODELNAME}/D_{MODELEPOCH}.pth /content/drive/MyDrive/{MODELNAME}_G_{MODELEPOCH}.pth\n",
@@ -188,7 +201,7 @@
         "#@markdown 模型名\n",
         "MODELNAME = \"lulu\"  #@param {type:\"string\"}\n",
         "#@markdown 模型epoch\n",
-        "MODELEPOCH = 6000  #@param {type:\"integer\"}\n",
+        "MODELEPOCH = 7500  #@param {type:\"integer\"}\n",
         "\n",
         "!mkdir -p /content/Retrieval-based-Voice-Conversion-WebUI/logs/{MODELNAME}\n",
         "\n",
@@ -241,7 +254,7 @@
         "#@markdown 模型名\n",
         "MODELNAME = \"lulu\"  #@param {type:\"string\"}\n",
         "#@markdown 停止的epoch\n",
-        "MODELEPOCH = 2500  #@param {type:\"integer\"}\n",
+        "MODELEPOCH = 3200  #@param {type:\"integer\"}\n",
         "#@markdown 保存epoch间隔\n",
         "EPOCHSAVE = 100  #@param {type:\"integer\"}\n",
         "#@markdown 采样率\n",
@@ -262,7 +275,7 @@
         "#@markdown 模型名\n",
         "MODELNAME = \"lulu\"  #@param {type:\"string\"}\n",
         "#@markdown 选中模型epoch\n",
-        "MODELEPOCH = 7700  #@param {type:\"integer\"}\n",
+        "MODELEPOCH = 9600  #@param {type:\"integer\"}\n",
         "\n",
         "!echo \"备份选中的模型。。。\"\n",
         "!cp /content/Retrieval-based-Voice-Conversion-WebUI/logs/{MODELNAME}/G_{MODELEPOCH}.pth /content/{MODELNAME}_D_{MODELEPOCH}.pth\n",
@@ -292,7 +305,7 @@
         "#@markdown 模型名\n",
         "MODELNAME = \"lulu\"  #@param {type:\"string\"}\n",
         "#@markdown 选中模型epoch\n",
-        "MODELEPOCH = 7700  #@param {type:\"integer\"}\n",
+        "MODELEPOCH = 9600  #@param {type:\"integer\"}\n",
         "\n",
         "!echo \"备份选中的模型。。。\"\n",
         "!cp /content/Retrieval-based-Voice-Conversion-WebUI/logs/{MODELNAME}/G_{MODELEPOCH}.pth /content/{MODELNAME}_D_{MODELEPOCH}.pth\n",

infer-web.py

@@ -9,7 +9,7 @@ import faiss
 ncpu=cpu_count()
 ngpu=torch.cuda.device_count()
 gpu_infos=[]
-if(torch.cuda.is_available()==False or ngpu==0):if_gpu_ok=False
+if((not torch.cuda.is_available()) or ngpu==0):if_gpu_ok=False
 else:
     if_gpu_ok = False
     for i in range(ngpu):
@@ -140,7 +140,7 @@ def uvr(model_name,inp_root,save_root_vocal,paths,save_root_ins):
         except:
             traceback.print_exc()
         print("clean_empty_cache")
-        torch.cuda.empty_cache()
+        if torch.cuda.is_available(): torch.cuda.empty_cache()
     yield "\n".join(infos)
 
 #一个选项卡全局只能有一个音色
@@ -152,7 +152,7 @@ def get_vc(sid):
             print("clean_empty_cache")
             del net_g, n_spk, vc, hubert_model,tgt_sr#,cpt
             hubert_model = net_g=n_spk=vc=hubert_model=tgt_sr=None
-            torch.cuda.empty_cache()
+            if torch.cuda.is_available(): torch.cuda.empty_cache()
             ###楼下不这么折腾清理不干净
             if_f0 = cpt.get("f0", 1)
             if (if_f0 == 1):
@@ -160,7 +160,7 @@ def get_vc(sid):
             else:
                 net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
             del net_g,cpt
-            torch.cuda.empty_cache()
+            if torch.cuda.is_available(): torch.cuda.empty_cache()
             cpt=None
         return {"visible": False, "__type__": "update"}
     person = "%s/%s" % (weight_root, sid)

infer/infer-pm-index256.py

@@ -104,7 +104,7 @@ for idx,name in enumerate(["冬之花clip1.wav",]):##
         "padding_mask": padding_mask.to(device),
         "output_layer": 9,  # layer 9
     }
-    torch.cuda.synchronize()
+    if torch.cuda.is_available(): torch.cuda.synchronize()
     t0=ttime()
     with torch.no_grad():
         logits = model.extract_features(**inputs)
@@ -116,13 +116,13 @@ for idx,name in enumerate(["冬之花clip1.wav",]):##
     feats = torch.from_numpy(big_npy[I.squeeze()].astype("float16")).unsqueeze(0).to(device)
 
     feats=F.interpolate(feats.permute(0,2,1),scale_factor=2).permute(0,2,1)
-    torch.cuda.synchronize()
+    if torch.cuda.is_available(): torch.cuda.synchronize()
     t1=ttime()
     # p_len = min(feats.shape[1],10000,pitch.shape[0])#太大了爆显存
     p_len = min(feats.shape[1],10000)#
     pitch, pitchf = get_f0(audio, p_len,f0_up_key)
     p_len = min(feats.shape[1],10000,pitch.shape[0])#太大了爆显存
-    torch.cuda.synchronize()
+    if torch.cuda.is_available(): torch.cuda.synchronize()
     t2=ttime()
     feats = feats[:,:p_len, :]
     pitch = pitch[:p_len]
@@ -133,7 +133,7 @@ for idx,name in enumerate(["冬之花clip1.wav",]):##
     pitchf = torch.FloatTensor(pitchf).unsqueeze(0).to(device)
     with torch.no_grad():
         audio = net_g.infer(feats, p_len,pitch,pitchf,sid)[0][0, 0].data.cpu().float().numpy()#nsf
-    torch.cuda.synchronize()
+    if torch.cuda.is_available(): torch.cuda.synchronize()
     t3=ttime()
     ta0+=(t1-t0)
     ta1+=(t2-t1)

requirements.txt

@@ -4,7 +4,7 @@ scipy==1.9.3
 librosa==0.9.2
 llvmlite==0.39.0
 fairseq==0.12.2
-faiss-cpu==1.7.2
+faiss-cpu==1.7.0
 gradio
 Cython
 future>=0.18.3

vc_infer_pipeline.py

@@ -72,6 +72,7 @@ class VC(object):
             "output_layer": 9,  # layer 9
         }
         t0 = ttime()
+        print("vc npy start time:", t0)
         with torch.no_grad():
             logits = model.extract_features(**inputs)
             feats  = model.final_proj(logits[0])
@@ -79,13 +80,14 @@ class VC(object):
         if(isinstance(index,type(None))==False and isinstance(big_npy,type(None))==False and index_rate!=0):
             npy = feats[0].cpu().numpy()
             if(self.is_half==True):npy=npy.astype("float32")
-            D, I = index.search(npy, 1)
+            _, I = index.search(npy, 1)
             npy=big_npy[I.squeeze()]
             if(self.is_half==True):npy=npy.astype("float16")
             feats = torch.from_numpy(npy).unsqueeze(0).to(self.device)*index_rate + (1-index_rate)*feats
 
         feats = F.interpolate(feats.permute(0, 2, 1), scale_factor=2).permute(0, 2, 1)
         t1 = ttime()
+        print("vc infer start time:", t1)
         p_len = audio0.shape[0]//self.window
         if(feats.shape[1]<p_len):
             p_len=feats.shape[1]
@@ -99,8 +101,9 @@ class VC(object):
             else:
                 audio1 = (net_g.infer(feats, p_len, sid)[0][0, 0] * 32768).data.cpu().float().numpy().astype(np.int16)
         del feats,p_len,padding_mask
-        torch.cuda.empty_cache()
+        if torch.cuda.is_available(): torch.cuda.empty_cache()
         t2 = ttime()
+        print("vc infer end time:", t2)
         times[0] += (t1 - t0)
         times[2] += (t2 - t1)
         return audio1
@@ -125,6 +128,7 @@ class VC(object):
         audio_opt=[]
         t=None
         t1=ttime()
+        print("f0 start time:", t1)
         audio_pad = np.pad(audio, (self.t_pad, self.t_pad), mode='reflect')
         p_len=audio_pad.shape[0]//self.window
         inp_f0=None
@@ -146,6 +150,7 @@ class VC(object):
             pitch = torch.tensor(pitch,device=self.device).unsqueeze(0).long()
             pitchf = torch.tensor(pitchf,device=self.device).unsqueeze(0).float()
         t2=ttime()
+        print("f0 end time:", t2)
         times[1] += (t2 - t1)
         for t in opt_ts:
             t=t//self.window*self.window
@@ -160,5 +165,5 @@ class VC(object):
             audio_opt.append(self.vc(model,net_g,sid,audio_pad[t:],None,None,times,index,big_npy,index_rate)[self.t_pad_tgt:-self.t_pad_tgt])
         audio_opt=np.concatenate(audio_opt)
         del pitch,pitchf,sid
-        torch.cuda.empty_cache()
+        if torch.cuda.is_available(): torch.cuda.empty_cache()
         return audio_opt