File size: 13,510 Bytes
e96a1ea 21bb99c e96a1ea |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 |
<div align="center">
<h1> Variational Inference with adversarial learning for end-to-end Singing Voice Conversion based on VITS </h1>
[![Hugging Face Spaces](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue)](https://huggingface.co/spaces/maxmax20160403/sovits5.0)
[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1PY1E4bDAeHbAD4r99D_oYXB46fG8nIA5?usp=sharing)
<img alt="GitHub Repo stars" src="https://img.shields.io/github/stars/PlayVoice/so-vits-svc-5.0">
<img alt="GitHub forks" src="https://img.shields.io/github/forks/PlayVoice/so-vits-svc-5.0">
<img alt="GitHub issues" src="https://img.shields.io/github/issues/PlayVoice/so-vits-svc-5.0">
<img alt="GitHub" src="https://img.shields.io/github/license/PlayVoice/so-vits-svc-5.0">
</div>
- πThis project is target for: beginners in deep learning, the basic operation of Python and PyTorch is the prerequisite for using this project;
- πThis project aims to help deep learning beginners get rid of boring pure theoretical learning, and master the basic knowledge of deep learning by combining it with practice;
- πThis project does not support real-time voice change; (support needs to replace whisper)
- πThis project will not develop one-click packages for other purposesοΌ
![sovits_framework](https://github.com/PlayVoice/so-vits-svc-5.0/assets/16432329/402cf58d-6d03-4d0b-9d6a-94f079898672)
- 6G memory GPU can be used to trained
- support for multiple speakers
- create unique speakers through speaker mixing
- even with light accompaniment can also be converted
- F0 can be edited using Excel
## Model properties
https://github.com/PlayVoice/so-vits-svc-5.0/releases/tag/hifigan_release
- [sovits5.0_main_1500.pth](https://github.com/PlayVoice/so-vits-svc-5.0/releases/download/hifigan_release/sovits5.0_main_1500.pth) The model includes: generator + discriminator = 176M, which can be used as a pre-training model
- speakers files are in the configs/singers directory, which can be used for reasoning tests, especially for timbre leakage
- speakers 22, 30, 47, and 51 are highly recognizable, and the training audio samples are in the configs/singers_sample directory
| Feature | From | Status | Function | Remarks |
| --- | --- | --- | --- | --- |
| whisper | OpenAI | β
| strong noise immunity | - |
| bigvgan | NVIDA | β
| alias and snake | The GPU takes up a little more, and the main branch is deleted; You need to switch to the branch [bigvgan](https://github.com/PlayVoice/so-vits-svc-5.0/tree/bigvgan)οΌthe formant is clearer and the sound quality is obviously improved |
| natural speech | Microsoft | β
| reduce mispronunciation | - |
| neural source-filter | NII | β
| solve the problem of audio F0 discontinuity | - |
| speaker encoder | Google | β
| Timbre Encoding and Clustering | - |
| GRL for speaker | Ubisoft |β
| Preventing Encoder Leakage Timbre | - |
| one shot vits | Samsung | β
| Voice Clone | - |
| SCLN | Microsoft | β
| Improve Clone | - |
| PPG perturbation | this project | β
| Improved noise immunity and de-timbre | - |
| VAE perturbation | this project | β
| Improve sound quality | - |
πdue to the use of data perturbation, it takes longer to train than other projects.
## Dataset preparation
Necessary pre-processing:
- 1 accompaniment separation
- 2 band extension
- 3 sound quality improvement
- 4 cut audio, less than 30 seconds for whisperπ
then put the dataset into the dataset_raw directory according to the following file structure
```shell
dataset_raw
ββββspeaker0
β ββββ000001.wav
β ββββ...
β ββββ000xxx.wav
ββββspeaker1
ββββ000001.wav
ββββ...
ββββ000xxx.wav
```
## Install dependencies
- 1 software dependency
> apt update && sudo apt install ffmpeg
> pip install -r requirements.txt
- 2 download the Timbre Encoder: [Speaker-Encoder by @mueller91](https://drive.google.com/drive/folders/15oeBYf6Qn1edONkVLXe82MzdIi3O_9m3), put `best_model.pth.tar` into `speaker_pretrain/`
- 3 download whisper model [multiple language medium model](https://openaipublic.azureedge.net/main/whisper/models/345ae4da62f9b3d59415adc60127b97c714f32e89e936602e85993674d08dcb1/medium.pt), Make sure to download `medium.pt`οΌput it into `whisper_pretrain/`
- 4 whisper is built-in, do not install it additionally, it will conflict and report an error
## Data preprocessing
- 1οΌ set working directory:
> export PYTHONPATH=$PWD
- 2οΌ re-sampling
generate audio with a sampling rate of 16000HzοΌ./data_svc/waves-16k
> python prepare/preprocess_a.py -w ./dataset_raw -o ./data_svc/waves-16k -s 16000
generate audio with a sampling rate of 32000HzοΌ./data_svc/waves-32k
> python prepare/preprocess_a.py -w ./dataset_raw -o ./data_svc/waves-32k -s 32000
- 3οΌ use 16K audio to extract pitchοΌf0_ceil=900, it needs to be modified according to the highest pitch of your data
> python prepare/preprocess_f0.py -w data_svc/waves-16k/ -p data_svc/pitch
or use next for low quality audio
> python prepare/preprocess_f0_crepe.py -w data_svc/waves-16k/ -p data_svc/pitch
- 4οΌ use 16K audio to extract ppg
> python prepare/preprocess_ppg.py -w data_svc/waves-16k/ -p data_svc/whisper
- 5οΌ use 16k audio to extract timbre code
> python prepare/preprocess_speaker.py data_svc/waves-16k/ data_svc/speaker
- 6οΌ extract the average value of the timbre code for inference; it can also replace a single audio timbre in generating the training index, and use it as the unified timbre of the speaker for training
> python prepare/preprocess_speaker_ave.py data_svc/speaker/ data_svc/singer
- 7οΌ use 32k audio to extract the linear spectrum
> python prepare/preprocess_spec.py -w data_svc/waves-32k/ -s data_svc/specs
- 8οΌ use 32k audio to generate training index
> python prepare/preprocess_train.py
- 9οΌ training file debugging
> python prepare/preprocess_zzz.py
```shell
data_svc/
βββ waves-16k
β βββ speaker0
β β βββ 000001.wav
β β βββ 000xxx.wav
β βββ speaker1
β βββ 000001.wav
β βββ 000xxx.wav
βββ waves-32k
β βββ speaker0
β β βββ 000001.wav
β β βββ 000xxx.wav
β βββ speaker1
β βββ 000001.wav
β βββ 000xxx.wav
βββ pitch
β βββ speaker0
β β βββ 000001.pit.npy
β β βββ 000xxx.pit.npy
β βββ speaker1
β βββ 000001.pit.npy
β βββ 000xxx.pit.npy
βββ whisper
β βββ speaker0
β β βββ 000001.ppg.npy
β β βββ 000xxx.ppg.npy
β βββ speaker1
β βββ 000001.ppg.npy
β βββ 000xxx.ppg.npy
βββ speaker
β βββ speaker0
β β βββ 000001.spk.npy
β β βββ 000xxx.spk.npy
β βββ speaker1
β βββ 000001.spk.npy
β βββ 000xxx.spk.npy
βββ singer
βββ speaker0.spk.npy
βββ speaker1.spk.npy
```
## Train
- 0οΌ if fine-tuning based on the pre-trained model, you need to download the pre-trained model: sovits5.0_main_1500.pth
> set pretrain: "./sovits5.0_main_1500.pth" in configs/base.yamlοΌand adjust the learning rate appropriately, eg 1e-5
- 1οΌ set working directory
> export PYTHONPATH=$PWD
- 2οΌ start training
> python svc_trainer.py -c configs/base.yaml -n sovits5.0
- 3οΌ resume training
> python svc_trainer.py -c configs/base.yaml -n sovits5.0 -p chkpt/sovits5.0/***.pth
- 4οΌ view log
> tensorboard --logdir logs/
![sovits5 0_base](https://github.com/PlayVoice/so-vits-svc-5.0/assets/16432329/1628e775-5888-4eac-b173-a28dca978faa)
## Inference
- 1οΌ set working directory
> export PYTHONPATH=$PWD
- 2οΌ export inference model: text encoder, Flow network, Decoder network
> python svc_export.py --config configs/base.yaml --checkpoint_path chkpt/sovits5.0/***.pt
- 3οΌ use whisper to extract content encoding, without using one-click reasoning, in order to reduce GPU memory usage
> python whisper/inference.py -w test.wav -p test.ppg.npy
generate test.ppg.npy; if no ppg file is specified in the next step, generate it automatically
- 4οΌ extract the F0 parameter to the csv text format, open the csv file in Excel, and manually modify the wrong F0 according to Audition or SonicVisualiser
> python pitch/inference.py -w test.wav -p test.csv
- 5οΌspecify parameters and infer
> python svc_inference.py --config configs/base.yaml --model sovits5.0.pth --spk ./configs/singers/singer0001.npy --wave test.wav --ppg test.ppg.npy --pit test.csv
when --ppg is specified, when the same audio is reasoned multiple times, it can avoid repeated extraction of audio content codes; if it is not specified, it will be automatically extracted;
when --pit is specified, the manually tuned F0 parameter can be loaded; if not specified, it will be automatically extracted;
generate files in the current directory:svc_out.wav
| args |--config | --model | --spk | --wave | --ppg | --pit | --shift |
| --- | --- | --- | --- | --- | --- | --- | --- |
| name | config path | model path | speaker | wave input | wave ppg | wave pitch | pitch shift |
## Creat singer
named by pure coincidenceοΌaverage -> ave -> evaοΌeve(eva) represents conception and reproduction
> python svc_eva.py
```python
eva_conf = {
'./configs/singers/singer0022.npy': 0,
'./configs/singers/singer0030.npy': 0,
'./configs/singers/singer0047.npy': 0.5,
'./configs/singers/singer0051.npy': 0.5,
}
```
the generated singer file isοΌeva.spk.npy
πboth Flow and Decoder need to input timbres, and you can even input different timbre parameters to the two modules to create more unique timbres.
## Data set
| Name | URL |
| --- | --- |
|KiSing |http://shijt.site/index.php/2021/05/16/kising-the-first-open-source-mandarin-singing-voice-synthesis-corpus/|
|PopCS |https://github.com/MoonInTheRiver/DiffSinger/blob/master/resources/apply_form.md|
|opencpop |https://wenet.org.cn/opencpop/download/|
|Multi-Singer |https://github.com/Multi-Singer/Multi-Singer.github.io|
|M4Singer |https://github.com/M4Singer/M4Singer/blob/master/apply_form.md|
|CSD |https://zenodo.org/record/4785016#.YxqrTbaOMU4|
|KSS |https://www.kaggle.com/datasets/bryanpark/korean-single-speaker-speech-dataset|
|JVS MuSic |https://sites.google.com/site/shinnosuketakamichi/research-topics/jvs_music|
|PJS |https://sites.google.com/site/shinnosuketakamichi/research-topics/pjs_corpus|
|JUST Song |https://sites.google.com/site/shinnosuketakamichi/publication/jsut-song|
|MUSDB18 |https://sigsep.github.io/datasets/musdb.html#musdb18-compressed-stems|
|DSD100 |https://sigsep.github.io/datasets/dsd100.html|
|Aishell-3 |http://www.aishelltech.com/aishell_3|
|VCTK |https://datashare.ed.ac.uk/handle/10283/2651|
## Code sources and references
https://github.com/facebookresearch/speech-resynthesis [paper](https://arxiv.org/abs/2104.00355)
https://github.com/jaywalnut310/vits [paper](https://arxiv.org/abs/2106.06103)
https://github.com/openai/whisper/ [paper](https://arxiv.org/abs/2212.04356)
https://github.com/NVIDIA/BigVGAN [paper](https://arxiv.org/abs/2206.04658)
https://github.com/mindslab-ai/univnet [paper](https://arxiv.org/abs/2106.07889)
https://github.com/nii-yamagishilab/project-NN-Pytorch-scripts/tree/master/project/01-nsf
https://github.com/brentspell/hifi-gan-bwe
https://github.com/mozilla/TTS
https://github.com/OlaWod/FreeVC [paper](https://arxiv.org/abs/2210.15418)
[SNAC : Speaker-normalized Affine Coupling Layer in Flow-based Architecture for Zero-Shot Multi-Speaker Text-to-Speech](https://github.com/hcy71o/SNAC)
[Adapter-Based Extension of Multi-Speaker Text-to-Speech Model for New Speakers](https://arxiv.org/abs/2211.00585)
[AdaSpeech: Adaptive Text to Speech for Custom Voice](https://arxiv.org/pdf/2103.00993.pdf)
[Cross-Speaker Prosody Transfer on Any Text for Expressive Speech Synthesis](https://github.com/ubisoft/ubisoft-laforge-daft-exprt)
[Learn to Sing by Listening: Building Controllable Virtual Singer by Unsupervised Learning from Voice Recordings](https://arxiv.org/abs/2305.05401)
[Adversarial Speaker Disentanglement Using Unannotated External Data for Self-supervised Representation Based Voice Conversion](https://arxiv.org/pdf/2305.09167.pdf)
[Speaker normalization (GRL) for self-supervised speech emotion recognition](https://arxiv.org/abs/2202.01252)
## Method of Preventing Timbre Leakage Based on Data Perturbation
https://github.com/auspicious3000/contentvec/blob/main/contentvec/data/audio/audio_utils_1.py
https://github.com/revsic/torch-nansy/blob/main/utils/augment/praat.py
https://github.com/revsic/torch-nansy/blob/main/utils/augment/peq.py
https://github.com/biggytruck/SpeechSplit2/blob/main/utils.py
https://github.com/OlaWod/FreeVC/blob/main/preprocess_sr.py
## Contributors
<a href="https://github.com/PlayVoice/so-vits-svc/graphs/contributors">
<img src="https://contrib.rocks/image?repo=PlayVoice/so-vits-svc" />
</a>
---
license: afl-3.0
--- |