Translation-Equivariant Self-Supervised Learning for Pitch Estimation with Optimal Transport

TL;DR

This paper introduces an optimal transport-based self-supervised learning method for pitch estimation that is translation-equivariant, offering a more stable and theoretically sound alternative to existing approaches.

Contribution

It presents a novel optimal transport objective for training translation-equivariant systems, specifically applied to pitch estimation, improving stability and theoretical grounding.

Findings

Enhanced numerical stability in pitch estimation models

Theoretically grounded training method

Simpler alternative to existing self-supervised approaches

Abstract

In this paper, we propose an Optimal Transport objective for learning one-dimensional translation-equivariant systems and demonstrate its applicability to single pitch estimation. Our method provides a theoretically grounded, more numerically stable, and simpler alternative for training state-of-the-art self-supervised pitch estimators.