Neural network AI DJ
Private service with neural network integration for repeating light shows
01 Task
The client was forced to manually press commands to activate effects during the performance, which distracted from the creative process, increased the risk of real-time errors and prevented focus on the show itself
02 Solution
A neural network has been developed in PyTorch that predicts the required commands in real time. The network “hears” the track and understands when and what effect is appropriate, no worse than an experienced DJ whose data was used for training
03 Solved tasks
- Designed an ML pipeline from data collection to real-time inference on PyTorch
- Trained the neural network to predict the desired effect (F1 > 0.8)
- Implemented streaming processing of a track with prediction during playback
- Integrated the model into DJ software and removed manual clicks
04 Result
Now the show is run by a neural network in a semi-automatic mode, and the DJ plays the role of a controller, focusing on creativity and interaction with the public, rather than on routine clicks. Technology paves the way for even greater automation of live shows in the future