EA-SVC
所属分类：模式识别(视觉/语音等)
开发工具：Python
文件大小：22KB
下载次数：0
上传日期：2020-11-04 03:27:49
上 传 者：
sh-1993
说明： “基于语音后验图的对抗性训练多对多歌声转换”的实现
(An implement of "Phonetic Posteriorgrams based Many-to-Many Singing Voice Conversion via Adversarial Training")
文件列表:
LICENSE (1067, 2020-11-04)
configs (0, 2020-11-04)
configs\e2e_stage1.json (1140, 2020-11-04)
configs\e2e_stage2.json (1532, 2020-11-04)
configs\e2e_stage3.json (1537, 2020-11-04)
distributed.py (7751, 2020-11-04)
inference.py (7251, 2020-11-04)
models (0, 2020-11-04)
models\disentangler.py (1180, 2020-11-04)
models\encoder.py (3379, 2020-11-04)
models\generator.py (1463, 2020-11-04)
models\modules.py (1394, 2020-11-04)
models\multiscale.py (2602, 2020-11-04)
train.py (12503, 2020-11-04)
utils (0, 2020-11-04)
utils\dataset.py (3641, 2020-11-04)
utils\loss.py (3293, 2020-11-04)
utils\optimizer.py (6806, 2020-11-04)
utils\utils.py (1469, 2020-11-04)
# EA-SVC
An implement of "Phonetic Posteriorgrams based Many-to-Many Singing Voice Conversion via Adversarial Training"
## Data prepare
1. PPG features (10ms frameshift)
2. F0 features (10ms frameshift)
3. Speaker embedding (One embedding per wav file)
4. Audio files (wave format, 24000 sample rate, mono)
## Write Configuration
Set path / directory or other configurations in .json files in directory "configs"
Rewrite your data load function in utils/dataset.py
## Model Training
Single GPU
```bash
CUDA_VISIBLE_DEVICES=0 python train.py -c configs/stage1.json
CUDA_VISIBLE_DEVICES=0 python train.py -c configs/stage2.json
CUDA_VISIBLE_DEVICES=0 python train.py -c configs/stage3.json
```
近期下载者：
相关文件：
收藏者：