MelBank

This project objective is to separate the sound of multiple speakers on a single channel.

It enables not only noise-speech separation, but also speech-speech separation.

Demo

Cannot play demo audio in GitHub. If you want to listen to demo audio, look this.

Requirement

• Python ~> 3.8
• TensorFlow

Installation

$ git clone <this repo>
$ cd <this repo>

$ pipenv install

You also need to install portaudio.

• macOS - brew install portaudio
• Ubuntu - sudo apt-get install portaudio19-dev

Usage

1. Create teacher data

$ pipenv run record # Recording each sound source to be separated
$ pipenv run build  # Build teacher data

2. Training

$ pipenv run train

3. Start demo!

$ pipenv run demo

If you want to know the details of how to use this, run the following command.

$ pipenv run help