![WaveGlow](waveglow_logo.png "WaveGLow")
## WaveGlow: a Flow-based Generative Network for Speech Synthesis
### Ryan Prenger, Rafael Valle, and Bryan Catanzaro
In our recent [paper], we propose WaveGlow: a flow-based network capable of
generating high quality speech from mel-spectrograms. WaveGlow combines insights
from [Glow] and [WaveNet] in order to provide fast, efficient and high-quality
audio synthesis, without the need for auto-regression. WaveGlow is implemented
using only a single network, trained using only a single cost function:
maximizing the likelihood of the training data, which makes the training
procedure simple and stable.
Our [PyTorch] implementation produces audio samples at a rate of 4850
kHz on an NVIDIA V100 GPU. Mean Opinion Scores show that it delivers audio
quality as good as the best publicly available WaveNet implementation.
Visit our [website] for audio samples.
## Setup
1. Clone our repo and initialize submodule
```command
git clone https://github.com/NVIDIA/waveglow.git
cd waveglow
git submodule init
git submodule update
```
2. Install requirements `pip3 install -r requirements.txt`
3. Install [Apex]
## Generate audio with our pre-existing model
1. Download our [published model]
2. Download [mel-spectrograms]
3. Generate audio `python3 inference.py -f <(ls mel_spectrograms/*.pt) -w waveglow_256channels.pt -o . --is_fp16 -s 0.6`
N.b. use `convert_model.py` to convert your older models to the current model
with fused residual and skip connections.
## Train your own model
1. Download [LJ Speech Data]. In this example it's in `data/`
2. Make a list of the file names to use for training/testing
```command
ls data/*.wav | tail -n+10 > train_files.txt
ls data/*.wav | head -n10 > test_files.txt
```
3. Train your WaveGlow networks
```command
mkdir checkpoints
python train.py -c config.json
```
For multi-GPU training replace `train.py` with `distributed.py`. Only tested with single node and NCCL.
For mixed precision training set `"fp16_run": true` on `config.json`.
4. Make test set mel-spectrograms
`python mel2samp.py -f test_files.txt -o . -c config.json`
5. Do inference with your network
```command
ls *.pt > mel_files.txt
python3 inference.py -f mel_files.txt -w checkpoints/waveglow_10000 -o . --is_fp16 -s 0.6
```
[//]: # (TODO)
[//]: # (PROVIDE INSTRUCTIONS FOR DOWNLOADING LJS)
[pytorch 1.0]: https://github.com/pytorch/pytorch#installation
[website]: https://nv-adlr.github.io/WaveGlow
[paper]: https://arxiv.org/abs/1811.00002
[WaveNet implementation]: https://github.com/r9y9/wavenet_vocoder
[Glow]: https://blog.openai.com/glow/
[WaveNet]: https://deepmind.com/blog/wavenet-generative-model-raw-audio/
[PyTorch]: http://pytorch.org
[published model]: https://ngc.nvidia.com/catalog/models/nvidia:waveglow_ljs_256channels
[mel-spectrograms]: https://drive.google.com/file/d/1g_VXK2lpP9J25dQFhQwx7doWl_p20fXA/view?usp=sharing
[LJ Speech Data]: https://keithito.com/LJ-Speech-Dataset
[Apex]: https://github.com/nvidia/apex
没有合适的资源?快使用搜索试试~ 我知道了~
基于Mellotron歌唱合成系统`内含预训练模型以及环境搭建教程可以直接推理使用.zip
共60个文件
py:28个
xml:11个
txt:7个
1.该资源内容由用户上传,如若侵权请联系客服进行举报
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
版权申诉
0 下载量 92 浏览量
2023-11-01
12:27:27
上传
评论
收藏 2.66MB ZIP 举报
温馨提示
基于Mellotron歌唱合成系统`内含预训练模型以及环境搭建教程可以直接推理使用 参考 mellotron 的 GST 部分,不存在 1:1 锁定。您可以像在其他存储库中一样使用 GST。 然而,如果您想使用 mellotron 模型进行推理,我们还从参考音频中提取两件事:节奏和音高,它们创建 1:1 对应关系。实际上是节奏创造了1:1的对应关系。但是,如果您不额外调节节奏,那么自动提取的音高可能没有意义。 如果您不需要节奏(您可以通过使用 model.inferece() 禁用)和音调调节(您可以通过发送零作为音调来禁用),您基本上会得到带有 GST 和扬声器 ID 的 tacotron 2。 ### 相关注释2 该论文指出,“目标说话者 St 总是可以在训练集中找到,而源文本、音高和节奏(Ts、Ps、Rs)可能来自训练集之外。”所以我认为源音频不需要扬声器 ID——对于训练集之外的某些任意输入音频来说,拥有有效的扬声器 ID 是没有意义的。然而,在examples_filelist.txt 中有一个发言者ID 列。这个专栏的意义何在? 该模型需要一个说话者 ID,因
资源推荐
资源详情
资源评论
收起资源包目录
基于Mellotron歌唱合成系统`内含预训练模型以及环境搭建教程可以直接推理使用.zip (60个子文件)
基于Mellotron歌唱合成系统:内含预训练模型以及环境搭建教程可以直接推理使用
mellotron
utils.py 1020B
data
cmu_dictionary 3.55MB
example2.wav 104KB
example1.wav 172KB
examples_filelist.txt 139B
yin.py 4KB
LICENSE 1KB
loss_scaler.py 4KB
hparams.py 4KB
layers.py 4KB
loss_function.py 673B
model.py 27KB
plotting_utils.py 2KB
mellotron_utils.py 16KB
text
__init__.py 3KB
LICENSE 1KB
numbers.py 2KB
cleaners.py 3KB
cmudict.py 2KB
symbols.py 812B
filelists
libritts_train_clean_100_audiopath_text_sid_shorterthan10s_atleast5min_train_filelist.txt 2.57MB
libritts_train_clean_100_audiopath_text_sid_atleast5min_val_filelist.txt 27KB
ljs_audiopaths_text_sid_val_filelist.txt 7KB
libritts_speakerinfo.txt 97KB
ljs_audiopaths_text_sid_train_filelist.txt 1.55MB
waveglow
denoiser.py 2KB
LICENSE 1KB
glow_old.py 9KB
distributed.py 7KB
mel2samp.py 6KB
glow.py 12KB
inference.py 4KB
convert_model.py 3KB
config.json 994B
README.md 3KB
modules.py 6KB
requirements.txt 155B
audio_processing.py 3KB
multiproc.py 647B
data_utils.py 6KB
stft.py 6KB
musicXML
last_voice_processed_3.xml 6KB
debussy_prelude_lyrics.musicxml 20KB
haendel_hallelujah.musicxml 288KB
last_voice_processed_6.xml 6KB
last_voice_processed_5.xml 6KB
test.xml 5KB
mozart_requiem_kyrie_satb.musicxml 517KB
last_voice_processed_2.xml 6KB
last_voice_processed.xml 4KB
haendel_hallelujah (copia).xml 288KB
song070_f00001_063.xml 24KB
song_text.xml 89KB
last_voice_processed_4.xml 10KB
trap.xml 166KB
Dockerfile 2KB
train_utils.py 5KB
checkpoints
README.md 137B
README.md 2KB
run_mellotron.py 6KB
共 60 条
- 1
资源评论
小码蚁.
- 粉丝: 2670
- 资源: 4523
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功