Voiceolation

Voiceolation is a music source separator that extracts vocals from songs. It is coded on Python and uses image segmentation methods via artificial intelligence, used a U-Net model implemented by Keras and Tensorflow.

Team Members

• Emir Yılmaz
• Kemalcan Güner
• Yunus Emre Günen
• Ahmad Amireh

Supervisor

Gökçe Nur Yılmaz [Scholar Link]

Jury Members

• Aslı Gençtav
• Venera Adanova

Download / Demo

You can isolate vocals and obtain as a .wav file if you have python. To extract vocals from any song: download voiceolation and install libraries on the requirements.txt.

1. Download and extract the .zip file
2. Install libraries from the requirements.txt
3. Run the below command giving a path of the sound file
py voiceolation.py path-to-soundfile

If you want more accurate result, change SR on the config.py because we are using downsampling.
P.S: This a student project, may not be satisfied but it doesn't hurt to give a chance and try :) [Download Link]

Demo

Dataset Samples

Original Input Sample	Downsampled Input	Output Audio File

Input Audio File	Output Audio File

Real-Life Samples

Input Audio File	Output Audio File

Spectrogram Example

Vocal Source of a Sample	Voiceolation Output of the Sample

TEDU CMPE 491 and CMPE 492 Reports

• [Project Proposal]
• [Project Specifications Report]
• [Analysis Report]
• [High-Level Design Report]
• [Low Level Design Report]
• [Test Plan Report]
• [Final Report]