Replicating band split RNN reveals energy costs in music separation
Investigating the Performance and Energy Costs of Replicating Band-Split RNN for Music Source Separation
Sound
Summary
Separating different sounds from music recordings is tricky, and one popular method uses a type of AI called band-split recurrent neural networks (BSRNN). The original method was partly hard to copy because its full code wasn’t shared. The paper recreates the entire process, tests different setups, and measures how much energy the training uses. The authors also share their code and models to help others avoid high energy costs and make such research easier to repeat.
What this means in practice
- •For audio engineers: Use the replicated band-split RNN model to separate vocals and instruments with known performance and energy requirements.
- •For machine learning practitioners: Access publicly released code and models to build music source separation systems while managing energy costs effectively.
Authors
Paul Magron, Romain Serizel, Constance Douwes
Abstract
Band-split recurrent neural network (BSRNN) is a popular music source separation model that yields close to state-of-the-art results using reasonable computational resources and public datasets. It is therefore interesting from a reproducible research perspective, but achieving its performance is not straightforward since its full code is not available. In this paper, we conduct a replication of BSRNN via implementing the full pipeline. We extend the original paper's analysis by experimentally studying various design choices about data preprocessing, the optimization protocol, and architectural parameters. We report and discuss this project's energy cost, and we underline how its footprint could have been substantial lower upon availability of the full pipeline, which advocates for more reproducible research practices. To comply with this objective, we publicly release our code and pre-trained models.