update README.md

README.md

@@ -1,12 +1,63 @@
 # SoftVC VITS Singing Voice Conversion
-## 重大BUG修复
-+ 断音问题已解决，音质提升了一个档次
-+ 2.0版本已经移至 sovits_2.0分支
-+ 3.0版本使用FreeVC的代码结构，与旧版本不通用
-
+## Update
+> 断音问题已解决，音质提升了一个档次\
+> 2.0版本已经移至 sovits_2.0分支\
+> 3.0版本使用FreeVC的代码结构，与旧版本不通用
 ## 模型简介
-歌声音色转换模型，通过SoftVC内容编码器提取源音频语音特征，与F0同时输入VITS替换原本的文本输入达到歌声转换的效果。
-> 目前模型修使用 [coarse F0](https://github.com/PlayVoice/VI-SVC/blob/main/svc/prepare/preprocess_wave.py) ，尝试使用[HarmoF0](https://github.com/wx-wei/harmof0) 进行f0提取但效果不佳，尝试使用[icassp2022-vocal-transcription](https://github.com/keums/icassp2022-vocal-transcription)提取midi替换f0输入但效果不佳
+歌声音色转换模型，通过SoftVC内容编码器提取源音频语音特征，与F0同时输入VITS替换原本的文本输入达到歌声转换的效果。同时，更换声码器为 [NSF HiFiGAN](https://github.com/openvpi/DiffSinger/tree/refactor/modules/nsf_hifigan) 解决断音问题
+
+## 数据集准备
+```shell
+仅需要以以下文件结构将数据集放入raw目录即可
+
+raw
+├───speaker0
+│   ├───xxx1-xxx1.wav
+│   ├───...
+│   └───Lxx-0xx8.wav
+└───speaker1
+    ├───xx2-0xxx2.wav
+    ├───...
+    └───xxx7-xxx007.wav
+```
+
+## 预先下载的模型文件
++ soft vc hubert：[hubert-soft-0d54a1f4.pt](https://github.com/bshall/hubert/releases/download/v0.1/hubert-soft-0d54a1f4.pt)
+  + 放在hubert目录下
++ 预训练模型文件 [G_0.pth D_0.pth](https://)
+  + 放在logs/48k 目录下
+  + 预训练模型为必选项，因为据测试从零开始训练有概率不收敛，同时也能加快训练速度
+  + 预训练模型删除了optimizer flow speakerembedding 等无关权重，因此可以认为基本剔除了旧的音色信息
+```shell
+# 一键下载
+# hubert
+wget -P hubert/ https://github.com/bshall/hubert/releases/download/v0.1/hubert-soft-0d54a1f4.pt
+# G与D预训练模型
+wget -P logs/48k/ https://
+wget -P logs/48k/ https://
+
+```
+## 数据预处理
+1. 重采样至 48khz
+
+```shell
+python resample.py
+ ```
+2. 自动划分训练集 验证集 测试集 以及配置文件
+```shell
+python preprocess_flist_config.py
+```
+3. 生成hubert与f0
+```shell
+python preprocess_hubert_f0.py
+```
+执行完以上步骤后 dataset 目录便是预处理完成的数据，可以删除raw文件夹了
+
+## 训练
+```shell
+python train.py -c configs/config.json -m 48k
+```
+
+## 推理
 
-模型推理、训练、一键脚本汇总整理仓库 [sovits_guide](https://github.com/IceKyrin/sovits_guide)
 

configs/config.json

@@ -4,9 +4,12 @@
     "eval_interval": 200,
     "seed": 1234,
     "epochs": 10000,
-    "learning_rate": 2e-4,
-    "betas": [0.8, 0.99],
-    "eps": 1e-9,
+    "learning_rate": 0.0002,
+    "betas": [
+      0.8,
+      0.99
+    ],
+    "eps": 1e-09,
     "batch_size": 16,
     "fp16_run": false,
     "lr_decay": 0.999875,
@@ -20,8 +23,8 @@
     "port": "8001"
   },
   "data": {
-    "training_files":"filelists/train.txt",
-    "validation_files":"filelists/val.txt",
+    "training_files": "filelists/train.txt",
+    "validation_files": "filelists/val.txt",
     "max_wav_value": 32768.0,
     "sampling_rate": 48000,
     "filter_length": 1280,
@@ -40,19 +43,49 @@
     "kernel_size": 3,
     "p_dropout": 0.1,
     "resblock": "1",
-    "resblock_kernel_sizes": [3,7,11],
-    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
-    "upsample_rates": [10,8,2,2],
+    "resblock_kernel_sizes": [
+      3,
+      7,
+      11
+    ],
+    "resblock_dilation_sizes": [
+      [
+        1,
+        3,
+        5
+      ],
+      [
+        1,
+        3,
+        5
+      ],
+      [
+        1,
+        3,
+        5
+      ]
+    ],
+    "upsample_rates": [
+      10,
+      8,
+      2,
+      2
+    ],
     "upsample_initial_channel": 512,
-    "upsample_kernel_sizes": [16,16,4,4],
+    "upsample_kernel_sizes": [
+      16,
+      16,
+      4,
+      4
+    ],
     "n_layers_q": 3,
     "use_spectral_norm": false,
     "gin_channels": 256,
-    "ssl_dim": 256
+    "ssl_dim": 256,
+    "n_speakers": 4
   },
-  "spk":{
-    "nen": 0,
-    "paimon": 1,
-    "yunhao": 2
+  "spk": {
+    "paimon": 0,
+    "nen": 1
   }
-}
+}

filelists/test.txt

@@ -1,12 +1,12 @@
-./dataset/48k/paimon/vo_ABLQ004_5_paimon_02.wav
-./dataset/48k/paimon/vo_ABLQ005_2_paimon_01.wav
-./dataset/48k/nen/kne110_005.wav
-./dataset/48k/paimon/vo_ABLQ004_6_paimon_02.wav
-./dataset/48k/paimon/vo_ABLQ004_6_paimon_01.wav
-./dataset/48k/nen/kne110_003.wav
-./dataset/48k/paimon/vo_ABLQ004_7_paimon_01.wav
-./dataset/48k/nen/kne110_004.wav
+./dataset/48k/nen/kne110_006.wav
 ./dataset/48k/paimon/vo_ABLQ005_2_paimon_02.wav
+./dataset/48k/paimon/vo_ABLQ005_2_paimon_01.wav
 ./dataset/48k/nen/kne110_001.wav
-./dataset/48k/nen/kne110_006.wav
+./dataset/48k/nen/kne110_003.wav
 ./dataset/48k/nen/kne110_002.wav
+./dataset/48k/paimon/vo_ABLQ004_6_paimon_01.wav
+./dataset/48k/paimon/vo_ABLQ004_6_paimon_02.wav
+./dataset/48k/paimon/vo_ABLQ004_5_paimon_02.wav
+./dataset/48k/nen/kne110_004.wav
+./dataset/48k/paimon/vo_ABLQ004_7_paimon_01.wav
+./dataset/48k/nen/kne110_005.wav

filelists/val.txt

@@ -1,4 +1,4 @@
-./dataset/48k/paimon/vo_ABLQ005_2_paimon_02.wav
+./dataset/48k/paimon/vo_ABLQ004_7_paimon_01.wav
 ./dataset/48k/nen/kne110_006.wav
-./dataset/48k/nen/kne110_002.wav
-./dataset/48k/paimon/vo_ABLQ004_5_paimon_02.wav
+./dataset/48k/paimon/vo_ABLQ005_2_paimon_02.wav
+./dataset/48k/nen/kne110_003.wav

models.py

@@ -280,6 +280,7 @@ def __init__(self,
     upsample_kernel_sizes,
     gin_channels,
     ssl_dim,
+    n_speakers,
     **kwargs):
 
     super().__init__()
@@ -300,7 +301,7 @@ def __init__(self,
     self.segment_size = segment_size
     self.gin_channels = gin_channels
     self.ssl_dim = ssl_dim
-    self.emb_g = nn.Embedding(10, gin_channels)
+    self.emb_g = nn.Embedding(n_speakers, gin_channels)
 
     self.enc_p_ = TextEncoder(ssl_dim, inter_channels, hidden_channels, 5, 1, 16,0, filter_channels, n_heads, p_dropout)
     hps = {

preprocess_flist_config.py

@@ -0,0 +1,117 @@
+import os
+import argparse
+from tqdm import tqdm
+from random import shuffle
+import json
+config_template = {
+  "train": {
+    "log_interval": 200,
+    "eval_interval": 200,
+    "seed": 1234,
+    "epochs": 10000,
+    "learning_rate": 2e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 16,
+    "fp16_run": False,
+    "lr_decay": 0.999875,
+    "segment_size": 17920,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0,
+    "use_sr": True,
+    "max_speclen": 384,
+    "port": "8001"
+  },
+  "data": {
+    "training_files":"filelists/train.txt",
+    "validation_files":"filelists/val.txt",
+    "max_wav_value": 32768.0,
+    "sampling_rate": 48000,
+    "filter_length": 1280,
+    "hop_length": 320,
+    "win_length": 1280,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": None
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [10,8,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4],
+    "n_layers_q": 3,
+    "use_spectral_norm": False,
+    "gin_channels": 256,
+    "ssl_dim": 256,
+    "n_speakers": 0,
+  },
+  "spk":{
+    "nen": 0,
+    "paimon": 1,
+    "yunhao": 2
+  }
+}
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--train_list", type=str, default="./filelists/train.txt", help="path to train list")
+    parser.add_argument("--val_list", type=str, default="./filelists/val.txt", help="path to val list")
+    parser.add_argument("--test_list", type=str, default="./filelists/test.txt", help="path to test list")
+    parser.add_argument("--source_dir", type=str, default="./dataset/48k", help="path to source dir")
+    args = parser.parse_args()
+
+    train = []
+    val = []
+    test = []
+    idx = 0
+    spk_dict = {}
+    spk_id = 0
+    for speaker in tqdm(os.listdir(args.source_dir)):
+        spk_dict[speaker] = spk_id
+        spk_id += 1
+        wavs = [os.path.join(args.source_dir, speaker, i)for i in os.listdir(os.path.join(args.source_dir, speaker))]
+        wavs = [i for i in wavs if i.endswith("wav")]
+        shuffle(wavs)
+        train += wavs[2:-10]
+        val += wavs[:2]
+        test += wavs[-10:]
+    n_speakers = len(spk_dict.keys())*2
+    shuffle(train)
+    shuffle(val)
+    shuffle(test)
+
+    print("Writing", args.train_list)
+    with open(args.train_list, "w") as f:
+        for fname in tqdm(train):
+            wavpath = fname
+            f.write(wavpath + "\n")
+
+    print("Writing", args.val_list)
+    with open(args.val_list, "w") as f:
+        for fname in tqdm(val):
+            wavpath = fname
+            f.write(wavpath + "\n")
+
+    print("Writing", args.test_list)
+    with open(args.test_list, "w") as f:
+        for fname in tqdm(test):
+            wavpath = fname
+            f.write(wavpath + "\n")
+
+    config_template["model"]["n_speakers"] = n_speakers
+    config_template["spk"] = spk_dict
+    print("Writing configs/config.json")
+    with open("configs/config.json", "w") as f:
+        json.dump(config_template, f, indent=2)

preprocess_hubert_f0.py

@@ -108,7 +108,7 @@ def process(filename):
         c = torch.load(save_name)
     f0path = filename+".f0.npy"
     if not os.path.exists(f0path):
-        cf0, f0 = compute_f0(filename, c.shape[-1] * 3)
+        cf0, f0 = get_f0(filename, c.shape[-1] * 3)
         np.save(f0path, f0)