AcapellaBot

Update [2020-05-20]: For better, modern acapella extraction / source separation, I would recommend Demucs, Spleeter, or PhonicMind (commercial product). This project worked reasonably well when I wrote it in 2017, but the current state of the art is much better :)

Original README continues below.

Isolating vocals from music with a Convolutional Neural Network. Blog post is here.

To Use:

1. Download the repo
2. Install the latest versions of Theano, Keras, librosa, and h5py on Python 3.
3. Set your dimension ordering to tf in ~/keras/keras.json:

   {
       "backend": "theano",
       "image_dim_ordering": "tf"
   }
   

4. Run python acapellabot.py song.mp3

Enjoy ❤