File size: 9,465 Bytes
5548515
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Accelerating Diffusion-based Singing Voice Conversion through Consistency Distillation
<br>
<div align="center">
<img src="../../../imgs/svc/DiffComoSVC.png" width="90%">
</div>
<br>

This is an implement of [Consistency Models](https://arxiv.org/abs/2303.01469) for accelerating diffusion-based singing voice conversion. The overall architecture follows "[Leveraging Content-based Features from Multiple Acoustic Models for Singing Voice Conversion](https://arxiv.org/abs/2310.11160)" (NeurIPS 2023 Workshop on Machine Learning for Audio), only a slightly modification is applied on acoustic model. Specifically,

* The acoustic model is a conformer which generates a coarse spectrogram and a diffusion decoder based on Bidirectional Non-Causal Dilated CNN which polish the former spectrogram for better. This is similar to [CoMoSpeech: One-Step Speech and Singing Voice Synthesis via Consistency Model](https://comospeech.github.io/)
* To accelerate diffusion model, we apply consistency distillation from [Consistency Models](https://arxiv.org/abs/2303.01469). For teacher model, the diffusion schedule of the diffusion decoder follows [karras diffusion](https://arxiv.org/abs/2206.00364). For distilling teacher model, the condition encoder and the conformer part of acoustic model are frozen while the diffusion decoder model is updated via exponential moving average. See Figure above for details.

There are five stages in total:

1. Data preparation
2. Features extraction
3. Teacher Model Training
4. Consistency Distillation
5. Inference/conversion

## 1. Data Preparation

### Dataset Download

By default, we utilize the five datasets for training: M4Singer, Opencpop, OpenSinger, SVCC, and VCTK. How to download them is detailed [here](../../datasets/README.md).

### Configuration

Specify the dataset paths in  `exp_config.json`. Note that you can change the `dataset` list to use your preferred datasets.

```json
    "dataset": [
        "m4singer",
        "opencpop",
        "opensinger",
        "svcc",
        "vctk"
    ],
    "dataset_path": {
        // TODO: Fill in your dataset path
        "m4singer": "[M4Singer dataset path]",
        "opencpop": "[Opencpop dataset path]",
        "opensinger": "[OpenSinger dataset path]",
        "svcc": "[SVCC dataset path]",
        "vctk": "[VCTK dataset path]"
    },
```

## 2. Features Extraction

### Content-based Pretrained Models Download

By default, we utilize the Whisper and ContentVec to extract content features. How to download them is detailed [here](../../../pretrained/README.md).

### Configuration

Specify the dataset path and the output path for saving the processed data and the training model in `exp_config.json`:

```json
    // TODO: Fill in the output log path
    "log_dir": "[Your path to save logs and checkpoints]",
    "preprocess": {
        // TODO: Fill in the output data path
        "processed_dir": "[Your path to save processed data]",
        ...
    },
```

### Run

Run the `run.sh` as the preproces stage (set  `--stage 1`).

```bash
cd Amphion
sh egs/svc/DiffComoSVC/run.sh --stage 1
```

Note: The `CUDA_VISIBLE_DEVICES` is set as `"0"` in default. You can change it when running `run.sh` by specifying such as `--gpu "1"`.

## 3. Teacher Model Training

### Configuration

Set the `distill` in `config/comosvc.json` to `false` for teacher model training, you can also specify the detailed configuration for conformer encoder and diffusion process here: 

```JSON
"comosvc":{
            "distill": false,
            // conformer encoder
            "input_dim": 384,
            "output_dim": 100,
            "n_heads": 2,
            "n_layers": 6,
            "filter_channels":512,
            // karras diffusion
            "P_mean": -1.2,
            "P_std": 1.2,
            "sigma_data": 0.5,
            "sigma_min": 0.002,
            "sigma_max": 80,
            "rho": 7,
            "n_timesteps": 40,
        },
```

We provide the default hyparameters in the `exp_config.json`. They can work on single NVIDIA-24g GPU. You can adjust them based on you GPU machines.

```json
"train": {
        "batch_size": 32,
        ...
        "adamw": {
            "lr": 2.0e-4
        },
        ...
    }
```

### Run

Run the `run.sh` as the training stage (set  `--stage 2`). Specify a experimental name to run the following command. The tensorboard logs and checkpoints will be saved in `[Your path to save logs and checkpoints]/[YourExptName]`.

```bash
cd Amphion
sh egs/svc/DiffComoSVC/run.sh --stage 2 --name [YourExptName]
```

Note: The `CUDA_VISIBLE_DEVICES` is set as `"0"` in default. You can specify it when running `run.sh` such as:

```bash
cd Amphion
sh egs/svc/DiffComoSVC/run.sh --stage 2 --name [YourExptName] --gpu "0,1,2,3"
```

## 4. Consistency Distillation

### Configuration

Set the `distill` in `config/comosvc.json` to `true` for teacher model training, and specify the `teacher_model_path` for consistency distillation. You can also specify the detailed configuration for conformer encoder and diffusion process here: 

```JSON
"model": {
    "teacher_model_path":"[Your_teacher_model_checkpoint].bin",
    ...
    "comosvc":{
                "distill": true,
                // conformer encoder
                "input_dim": 384,
                "output_dim": 100,
                "n_heads": 2,
                "n_layers": 6,
                "filter_channels":512,
                // karras diffusion
                "P_mean": -1.2,
                "P_std": 1.2,
                "sigma_data": 0.5,
                "sigma_min": 0.002,
                "sigma_max": 80,
                "rho": 7,
                "n_timesteps": 40,
            },
```

We provide the default hyparameters in the `exp_config.json`. They can work on single NVIDIA-24g GPU. You can adjust them based on you GPU machines.

```json
"train": {
        "batch_size": 32,
        ...
        "adamw": {
            "lr": 2.0e-4
        },
        ...
    }
```

### Run

Run the `run.sh` as the training stage (set  `--stage 2`). Specify a experimental name to run the following command. The tensorboard logs and checkpoints will be saved in `[Your path to save logs and checkpoints]/[YourExptName]`.

```bash
cd Amphion
sh egs/svc/DiffComoSVC/run.sh --stage 2 --name [YourExptName]
```

Note: The `CUDA_VISIBLE_DEVICES` is set as `"0"` in default. You can specify it when running `run.sh` such as:

```bash
cd Amphion
sh egs/svc/DiffComoSVC/run.sh --stage 2 --name [YourExptName] --gpu "0,1,2,3"
```

## 5. Inference/Conversion

### Pretrained Vocoder Download

We fine-tune the official BigVGAN pretrained model with over 120 hours singing voice data. The benifits of fine-tuning has been investigated in our paper (see this [demo page](https://www.zhangxueyao.com/data/MultipleContentsSVC/vocoder.html)). The final pretrained singing voice vocoder is released [here](../../../pretrained/README.md#amphion-singing-bigvgan) (called `Amphion Singing BigVGAN`).

### Run

For inference/conversion, you need to specify the following configurations when running `run.sh`:

| Parameters                                          | Description                                                  | Example                                                      |
| --------------------------------------------------- | ------------------------------------------------------------ | ------------------------------------------------------------ |
| `--infer_expt_dir`                                  | The experimental directory which contains `checkpoint`       | `[Your path to save logs and checkpoints]/[YourExptName]`    |
| `--infer_output_dir`                                | The output directory to save inferred audios.                | `[Your path to save logs and checkpoints]/[YourExptName]/result` |
| `--infer_source_file` or `--infer_source_audio_dir` | The inference source (can be a json file or a dir).          | The `infer_source_file` could be `[Your path to save processed data]/[YourDataset]/test.json`, and the `infer_source_audio_dir` is a folder which includes several audio files (*.wav, *.mp3 or *.flac). |
| `--infer_target_speaker`                            | The target speaker you want to convert into. You can refer to `[Your path to save logs and checkpoints]/[YourExptName]/singers.json` to choose a trained speaker. | For opencpop dataset, the speaker name would be `opencpop_female1`. |
| `--infer_key_shift`                                 | How many semitones you want to transpose.                    | `"autoshfit"` (by default), `3`, `-3`, etc.                  |

For example, if you want to make `opencpop_female1` sing the songs in the `[Your Audios Folder]`, just run:

```bash
cd Amphion
sh egs/svc/DiffComoSVC/run.sh --stage 3 --gpu "0" \
    --infer_expt_dir [Your path to save logs and checkpoints]/[YourExptName] \
    --infer_output_dir [Your path to save logs and checkpoints]/[YourExptName]/result \
    --infer_source_audio_dir [Your Audios Folder] \
    --infer_target_speaker "opencpop_female1" \
    --infer_key_shift "autoshift"
```
Specially, you can configurate the inference steps for teacher model by setting `inference` at `exp_config`(student model is always one-step sampling):
```json
    "inference": {
        "comosvc": {
            "inference_steps": 40
        }
    }
```

# Reference
https://github.com/zhenye234/CoMoSpeech

https://github.com/openai/consistency_models