This is a project for KAIST CS470: Introduction to A.I.
We implement singer-conversion. On the web page, users can convert the song with the singer they want.
Conversion Deep Learning Model: https://github.com/seo3650/Audio_style_transfer/.
On the page, you need 2 songs: the song you want to convert, and the song which the singer you want to sing. The songs should be mp3 (less than 5MB) or wav(less than 25MB) file. Also, the song's name should be in English with no space. (ex: abc.mp3 (O), ab_c.mp3 (O), ab c.mp3 (X))
After selecting the file and click the convert, wait ~5 minutes for conversion. If the conversion end, a converted song will be downloaded automatically.
- Python 3.6
- Pytorch 1.7.0
- pyworld
- tqdm
- librosa
- tensorboardX and tensorboard
We use StarGAN pretrained model. Please refer to: https://github.com/seo3650/Audio_style_transfer/
index.py is the main file. When a user requests a conversion, the index.py file receives the request. The conversion pipeline is below.
- Upload the user's mp3 files into the server
- Extract vocal & MR wav files from each of the 2 mp3 files
- Convert the voice of source wav file
- Merge the converted file with MR
This is the main page of our website.
Choose mp3 file of song and singer.
File name should not contain spaces and file name should be in English.
Our website will convert a singer of the song and the song will be automatically downloaded. It will take ~5 minutes.
(Too many people can be unstable)
- antd
- styled-components
- react-router-dom
- axios
Google Cloud Platform