A Simple HMM with Self-Supervised Representations for Phone Segmentation

TL;DR

This paper demonstrates that peak detection on Mel spectrograms is a strong baseline for phonetic segmentation, and introduces a simple HMM using self-supervised representations and boundary features that outperforms previous methods.

Contribution

The paper shows that a simple peak detection baseline can outperform many self-supervised methods and proposes a straightforward HMM approach leveraging self-supervised features for improved segmentation.

Findings

Peak detection on Mel spectrograms is a strong baseline.

The proposed HMM with self-supervised features improves segmentation results.

The approach offers a versatile and generalized framework.

Abstract

Despite the recent advance in self-supervised representations, unsupervised phonetic segmentation remains challenging. Most approaches focus on improving phonetic representations with self-supervised learning, with the hope that the improvement can transfer to phonetic segmentation. In this paper, contrary to recent approaches, we show that peak detection on Mel spectrograms is a strong baseline, better than many self-supervised approaches. Based on this finding, we propose a simple hidden Markov model that uses self-supervised representations and features at the boundaries for phone segmentation. Our results demonstrate consistent improvements over previous approaches, with a generalized formulation allowing versatile design adaptations.