Adversarial Unsupervised Domain Adaptation for Harmonic-Percussive Source Separation

TL;DR

This paper proposes an adversarial unsupervised domain adaptation method for harmonic-percussive source separation, enabling improved performance on new music domains without requiring labeled data, and introduces the Tap & Fiddle dataset.

Contribution

It introduces a novel adversarial unsupervised domain adaptation framework for music source separation and presents a new Scandinavian fiddle dataset with isolated tracks.

Findings

Improved separation performance on target domain without losing original domain accuracy

Effective adaptation using only unlabelled mixture data from target domain

Introduction of the Tap & Fiddle dataset for music research

Abstract

This paper addresses the problem of domain adaptation for the task of music source separation. Using datasets from two different domains, we compare the performance of a deep learning-based harmonic-percussive source separation model under different training scenarios, including supervised joint training using data from both domains and pre-training in one domain with fine-tuning in another. We propose an adversarial unsupervised domain adaptation approach suitable for the case where no labelled data (ground-truth source signals) from a target domain is available. By leveraging unlabelled data (only mixtures) from this domain, experiments show that our framework can improve separation performance on the new domain without losing any considerable performance on the original domain. The paper also introduces the Tap & Fiddle dataset, a dataset containing recordings of Scandinavian fiddle tunes along with isolated tracks for 'foot-tapping' and 'violin'.