Automatic Live Music Soundchecking with Reference Audio on a Laptop

Matthew Keating; Michael Casey

Automatic Live Music Soundchecking with Reference Audio on a Laptop
Image credit: Matthew Keating; Michael Casey

Abstract:

We introduce a live mixing system that gives visual recommendations directly to a guitarist while they play with a band, based on a reference track. Our work runs on a laptop computer and receives audio input through the laptop’s microphone. We call this system SoLAR: Soundchecking for Live Band Audio with Reference. This work focuses on guitarist loudness control, but the framework is built to scale to other instruments and control parameters. SoLAR uses modern source separation techniques to get instrument stems from the reference audio. We introduce a perceptual algorithm for calculating which part of the bark scale each instrument dominates to find a mix summarization. We fit a neural network model to predict the reference mix summarization. SoLAR makes recommendations for loudness adjustment by comparing the predicted reference mix to the current live mix in a policy function. We demonstrate that these recommendations improve on random guessing through simulations in a gymnasium environment. We describe the SoLAR graphical user interface and our design considerations. Then, we discuss a small formative study with guitarists. This study allowed us to iterate on the interface and showed promising early results for live mixing.