You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
lipku 9ad00be5cd update readme 7 months ago
assets update readme 1 year ago
data add wav2lip customvideo 12 months ago
ernerf resolve building errors 11 months ago
llm add wav2lip customvideo 12 months ago
models add musetalk 1 year ago
musetalk cuda memory does not increase with the number of concurrency 8 months ago
wav2lip cuda memory does not increase with the number of concurrency 8 months ago
web add wav2lip and musetalk model warmup 7 months ago
.gitignore add 1 year ago
Dockerfile add docker 2 years ago
LICENSE fix jquery.js 9 months ago
README.md update readme 7 months ago
app.py add wav2lip and musetalk model warmup 7 months ago
baseasr.py Optimize multiple sessions 8 months ago
basereal.py Optimize multiple sessions 8 months ago
lipasr.py cuda memory does not increase with the number of concurrency 8 months ago
lipreal.py add wav2lip and musetalk model warmup 7 months ago
museasr.py cuda memory does not increase with the number of concurrency 8 months ago
musereal.py add wav2lip and musetalk model warmup 7 months ago
nerfasr.py cuda memory does not increase with the number of concurrency 8 months ago
nerfreal.py fix: NameError in load_avatar 7 months ago
requirements.txt fix edgetts 8 months ago
ttsreal.py Optimize multiple sessions 8 months ago
webrtc.py cuda memory does not increase with the number of concurrency 8 months ago

README.md

Real time interactive streaming digital human realize audio video synchronous dialogue. It can basically achieve commercial effects.
实时交互流式数字人,实现音视频同步对话。基本可以达到商用效果

ernerf效果 musetalk效果 wav2lip效果

为避免与3d数字人混淆原项目metahuman-stream改名为livetalking原有链接地址继续可用

News

  • 2024.12.8 完善多并发,显存不随并发数增加
  • 2024.12.21 添加wav2lip、musetalk模型预热解决第一次推理卡顿问题

Features

  1. 支持多种数字人模型: ernerf、musetalk、wav2lip
  2. 支持声音克隆
  3. 支持数字人说话被打断
  4. 支持全身视频拼接
  5. 支持rtmp和webrtc
  6. 支持视频编排:不说话时播放自定义视频
  7. 支持多并发

1. Installation

Tested on Ubuntu 20.04, Python3.10, Pytorch 1.12 and CUDA 11.3

1.1 Install dependency

conda create -n nerfstream python=3.10
conda activate nerfstream
#如果cuda版本不为11.3(运行nvidia-smi确认版本),根据<https://pytorch.org/get-started/previous-versions/>安装对应版本的pytorch 
conda install pytorch==1.12.1 torchvision==0.13.1 cudatoolkit=11.3 -c pytorch
pip install -r requirements.txt
#如果不训练ernerf模型不需要安装下面的库
pip install "git+https://github.com/facebookresearch/pytorch3d.git"
pip install tensorflow-gpu==2.8.0
pip install --upgrade "protobuf<=3.20.1"

安装常见问题FAQ
linux cuda环境搭建可以参考这篇文章 https://zhuanlan.zhihu.com/p/674972886

2. Quick Start

默认采用ernerf模型webrtc推流到srs

2.1 运行srs

export CANDIDATE='<服务器外网ip>'  #如果srs与浏览器访问在同一层级内网不需要执行这步
docker run --rm --env CANDIDATE=$CANDIDATE \
  -p 1935:1935 -p 8080:8080 -p 1985:1985 -p 8000:8000/udp \
  registry.cn-hangzhou.aliyuncs.com/ossrs/srs:5 \
  objs/srs -c conf/rtc.conf

2.2 启动数字人:

python app.py

如果访问不了huggingface在运行前

export HF_ENDPOINT=https://hf-mirror.com

用浏览器打开http://serverip:8010/rtcpushapi.html, 在文本框输入任意文字,提交。数字人播报该段文字
备注:服务端需要开放端口 tcp:8000,8010,1985; udp:8000

3. More Usage

使用说明: https://livetalking-doc.readthedocs.io/

4. Docker Run

不需要前面的安装,直接运行。

docker run --gpus all -it --network=host --rm registry.cn-beijing.aliyuncs.com/codewithgpu2/lipku-metahuman-stream:vjo1Y6NJ3N

代码在/root/metahuman-stream先git pull拉一下最新代码然后执行命令同第2、3步

提供如下镜像

5. 性能分析

  1. 帧率
    在Tesla T4显卡上测试整体fps为18左右如果去掉音视频编码推流帧率在20左右。用4090显卡可以达到40多帧/秒。
  2. 延时
    整体延时3s左右
    1tts延时1.7s左右目前用的edgetts需要将每句话转完后一次性输入可以优化tts改成流式输入
    2wav2vec延时0.4s需要缓存18帧音频做计算 3srs转发延时设置srs服务器减少缓冲延时。具体配置可看 https://ossrs.net/lts/zh-cn/docs/v5/doc/low-latency

6. TODO

  • 添加chatgpt实现数字人对话
  • 声音克隆
  • 数字人静音时用一段视频代替
  • MuseTalk
  • Wav2Lip

如果本项目对你有帮助帮忙点个star。也欢迎感兴趣的朋友一起来完善该项目.

  • 知识星球: https://t.zsxq.com/7NMyO 沉淀高质量常见问题、最佳实践经验、问题解答
  • 微信公众号:数字人技术