Skip to the content.

This is the demo page for the paper “Automatic DJ Transitions with Differentiable Audio Effects and Generative Adversarial Networks”.

Abstract

A central task of a Disc Jockey (DJ) is to create a mixset of music with seamless transitions between adjacent tracks. In this paper, we explore a data-driven approach that uses a generative adversarial network to create the song transition by learning from real-world DJ mixes. In particular, the generator of the model uses two differentiable digital signal processing components, an Equalizer (EQ) and fader, to mix two tracks selected by a data generation pipeline. The generator has to set the parameters of the EQs and fader in such a way that the resulting mix resembles real mixes created by human DJ, as judged by the discriminator counterpart. Result of a listening test shows that the model can achieve competitive results compared with baselines.

Audio Samples


We include 8 groups (A–H) of paired tracks (x1, x2) from the testing set, and for each transition type, we include 2 groups of paired tracks. Furthermore, we create 5 mixes (x3^) in each row by the following approaches with the same paired. In colclusion, we provide totally 7 audios in each row which include 2 paired tracks and 5 mixs, the detailed is shown below.


Mixing Pipeline

  1. Prev : track which is considered to be the first track in a mix as shown in figure above (x1)
  2. Next : track which is considered to be the second track in a mix as shown in figure above (x2)
  3. Sum : transitions created by summation of two tracks without any effects
  4. Linear : transitions created by applying linear cross-fading in the transition region
  5. Rule : transitions created by general purpose expert DJs’ rule
  6. GAN : transitions created by GANs’ generator (our proposed)
  7. Human : transitions created by an expert DJ

non-vocal to non-vocal (nv-nv)

Prev Next Sum Linear Rule GAN Human
A
B

non-vocal to vocal (nv-v)

Prev Next Sum Linear Rule GAN Human
C
D

vocal to non-vocal (v-nv)

Prev Next Sum Linear Rule GAN Human
E
F

vocal to vocal (v-v)

Prev Next Sum Linear Rule GAN Human
G
H

Contact


Bo-Yu Chen: bernie40916@gmail.com