A neural network that divides a song into vocals and accompaniment

Then the guys from Deezer shared a neuron that is able to divide one song into components - separately vocals, separately instruments.

deezer / spleeterDeezer source separation library including pretrained models. – deezer / spleeter GITHUB.COM

Moreover, one song can be divided into 5 components: vocals, bass, drums, piano and everything else.

It works something like this, you give her great hits at the entrance:

The neuron, it seems to me, can be of interest to real musicians in the first place – if you want to pick it yourself, then you need to open this collaboration:

Google colaboratoryGoogle colaboratory COLAB.RESEARCH.GOOGLE.COM

And change the link to your mp3 file, also do not forget to change the names of mp3 files that are fed to the input of a neuron.

Back to top button