Deep Fake Audio VIdeo with colab

Contents:

There are three process

• Video Generation - A single image is enough
• Audio Generation - A single audio is enough you can make any words with it
• Combine Audio and video

Will covert all manual process to code soon

Entire process of audio and video generation is in colab so no need to install any thing in local . if you are in linux machine you can use "ffmeg command" to do all preprocessing and post processing below.

Video Generation

1. have a png file (target person)

2. have a video file (specimen person)

Execute the colab file for video

3. Use increasefps.py file to increase FPS and make your video better

How to generate good deep fake video

Use a 256x256 file and choose one png file in which it is till neck example see this drive (https://drive.google.com/drive/folders/1kZ1gCnpfU0BnpdU47pLM_TQ6RypDDqgw)

Audio Generation

Note it works for both male and female

Instructions in colab for good audio

please add within 12 words .. and if u want to generate more paragraph save the generated audio in another name and then generate another audio

Dont use any punctuations it will produce some noise . Use only one space between words

Have the any audio(.wav) file of the person who is the target and read the instruction below and then go to colab

if u dont have (.wav) follow below steps to generate quickly

Convert video to audio

1. find a youtube video where the target person is speaking(only the person should speak no background audio or else no other persons voice) and convert to mp3 use this site for converting https://ytmp3.cc/en13/ and download the video

Convert mp3 to wav

In linux - Using ffmpeg

ffmpeg -i input.mp3 output.wav 

or else follow 2 and 3 points

2. Open this site https://online-audio-converter.com/ and Upload the mp3 file and give a DVD resolution and click convert . Now hover over the "Download" button and copy the link . go to google colab and use below command to copy it to google colab

!wget <link>

3.Rename the file name to audio.wav

!mv <filename.wav> audio.wav

Audio merger : https://clideo.com/merge-wav

Sample audios that i used

(In the entire audio only the targeted person will be speaking )

• Steve jobs - https://www.youtube.com/watch?v=65_PmYipnpk
• Rohan Shravan - https://www.youtube.com/watch?v=PstVDlK2lGA
• Ritika sighn - https://www.youtube.com/watch?v=QQI7foZMPtE

Combine Audio and video

if you are in linux you can use below command

ffmpeg -i "mres/g_wonder.mp4.mp4" -i "properaudio/c_wonderwoman.wav" -shortest g_wonder.mp4

or else you can use below website which generates without water mark

Use this website - https://www.kapwing.com/tools/add-music-to-video

Will upgrade the code and process in manner and quality of audio and video asap

Credits :

Video - https://towardsdatascience.com/how-to-produce-a-deepfake-video-in-5-minutes-513984fd24b6

Video - https://www.youtube.com/watch?v=peOKeRBU_uQ

Audio - https://github.com/CorentinJ/Real-Time-Voice-Cloning.git