刚刚，我用三行代码创建了一个虚拟主播

刚刚，我花了10分钟，写了三行代码创建一个具有明星脸的虚拟主播

先看看效果：

语音播报虚拟主播2

实现简易的虚拟数字人非常简单，需要调用三个模型：

（1）First Order Motion（表情迁移）

（2）Text to Speech（文本转语音）

（2）Wav2Lip（唇形合成）。
具体技术步骤如下：
1，把图像放入First Order Motion模型进行面部表情迁移，让虚拟主播的表情更加逼近真人，既然定位是一个主播，那表情都参考当然是要用“国家级标准”的，所以参考的对象选择了梓萌老师~
2，通过Text to Speech模型，将输入的文字转换成音频输出。
3，得到面部表情迁移的视频和音频之后，通过Wav2Lip模型，将音频和视频合并，并根据音频内容调整唇形，使得虚拟人更加接近真人效果。

1、运行依赖安装

In [ ]

# 升级PaddleHub
!pip install --upgrade paddlehub

In [ ]

# 下载nltk_data
!wget https://paddlespeech.bj.bcebos.com/Parakeet/tools/nltk_data.tar.gz
!tar zxvf nltk_data.tar.gz

In [ ]

# 安装ParaKeet
%cd Parakeet/
!pip install -e.
%cd ..

In [ ]

# 安装依赖
!hub install first_order_motion==1.0.0
!hub install wav2lip
!hub install fastspeech2_baker==1.0.0

Download https://bj.bcebos.com/paddlehub/paddlehub_dev/first_order_motion.tar.gz
[##################################################] 100.00%
Decompress /home/aistudio/.paddlehub/tmp/tmphcuxe0xl/first_order_motion.tar.gz
[##################################################] 100.00%
[2022-01-07 15:32:45,388] [    INFO] - Installing dependent packages from /home/aistudio/.paddlehub/tmp/tmpvtfv5cjp/first_order_motion/requirements.txt: /

2、 开始创建虚拟人

2.1 表情驱动

通过FOM模型，输入图像和驱动视频，让人像动起来！

In [6]

import cv2
import os

files = os.listdir('input_data/img/')

for f in files:
    img  =  cv2.imread('input_data/img/'+f)
    imgshape = img.shape
    resimg = cv2.resize(img,(int(img.shape[1]/2),int(img.shape[0]/2)))
    cv2.imwrite('input_data/'+f,resimg)

In [1]

import paddlehub as hub

FOM_Module = hub.Module(name="first_order_motion")
FOM_Module.generate(source_image="input_data/t5.jpeg", # 输入图像
                    driving_video="input_data/zimeng.mp4", # 输入驱动视频
                    ratio=0.4, 
                    image_size=256, 
                    output_dir='./output/', # 输出文件夹
                    filename='FOM.mp4', # 输出文件名
                    use_gpu=True)

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/__init__.py:107: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working
  from collections import MutableMapping
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/rcsetup.py:20: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working
  from collections import Iterable, Mapping
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/colors.py:53: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working
  from collections import Sized
W0108 00:26:51.097970 19449 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.0, Runtime API Version: 10.1
W0108 00:26:51.104054 19449 device_context.cc:465] device: 0, cuDNN Version: 7.6.

[01/08 00:26:56] ppgan INFO: Found /home/aistudio/.cache/ppgan/GPEN-512.pdparams
1 persons have been detected

100%|██████████| 300/300 [00:36<00:00,  8.31it/s]

2.2 文本转语音

输入你想让虚拟数字人说的话，转换生成一段音频。

In [4]

sentences = ['开发者你好，欢迎使用飞桨，我是你的专属虚拟人。'] # 输入说话内容

TTS_Module = hub.Module(
    name='fastspeech2_baker',
    version='1.0.0')
wav_files =  TTS_Module.generate(sentences)
print(f'声音已生成，音频文件输出在{wav_files}')

[2022-01-07 15:48:35,288] [    INFO] - Load fastspeech2 params from /home/aistudio/.paddlehub/modules/fastspeech2_baker/assets/fastspeech2_nosil_baker_ckpt_0.4/snapshot_iter_76000.pdz
[2022-01-07 15:48:35,671] [    INFO] - Load vocoder params from /home/aistudio/.paddlehub/modules/fastspeech2_baker/assets/pwg_baker_ckpt_0.4/pwg_snapshot_iter_400000.pdz
Building prefix dict from the default dictionary ...
[2022-01-07 15:48:35] [DEBUG] [__init__.py:113] Building prefix dict from the default dictionary ...
Dumping model to file cache /tmp/jieba.cache
[2022-01-07 15:48:36] [DEBUG] [__init__.py:147] Dumping model to file cache /tmp/jieba.cache
Loading model cost 0.791 seconds.
[2022-01-07 15:48:36] [DEBUG] [__init__.py:165] Loading model cost 0.791 seconds.
Prefix dict has been built successfully.
[2022-01-07 15:48:36] [DEBUG] [__init__.py:166] Prefix dict has been built successfully.
[2022-01-07 15:48:48,064] [    INFO] - 1 wave files have been generated in /home/aistudio/wavs

声音已生成，音频文件输出在['/home/aistudio/wavs/1.wav']

2.3 唇形合成

把刚刚得到的动态视频和音频文件输入到Wav2Lip模型中，让唇形根据说话的内容动态改变。

In [2]

import paddlehub as hub
W2F_Module = hub.Module(name="wav2lip")

W2F_Module.wav2lip_transfer(face='output/FOM.mp4', 
                            audio='wavs/bo.wav', 
                            output_dir='./transfer_result/', 
                            use_gpu=True) 

 

虚拟的主播的图片可以随意更改，这里随便百度了两张明星图片，原图就不放了，看下一些效果

最后效果：

效果1：

语音播报虚拟主播1

效果2：

语音播报虚拟主播2

 

AI Stdio 可直接运行，欢迎fork paddle虚拟数字主播播新闻 - 飞桨AI Studio - 人工智能学习与实训社区