Self-supervised Predictive Coding Models Encode Speaker and Phonetic Information in Orthogonal Subspaces

TL;DR

This paper demonstrates that self-supervised speech models encode speaker and phonetic information in nearly orthogonal subspaces, enabling effective speaker normalization without transcriptions and improving phonetic discrimination.

Contribution

The study reveals the orthogonal subspace structure of speaker and phonetic information in self-supervised models and introduces a novel normalization method leveraging this property.

Findings

Orthogonal subspaces for speaker and phonetic info identified via PCA

Proposed normalization method effectively removes speaker info

Method outperforms previous baselines in phone discrimination

Abstract

Self-supervised speech representations are known to encode both speaker and phonetic information, but how they are distributed in the high-dimensional space remains largely unexplored. We hypothesize that they are encoded in orthogonal subspaces, a property that lends itself to simple disentanglement. Applying principal component analysis to representations of two predictive coding models, we identify two subspaces that capture speaker and phonetic variances, and confirm that they are nearly orthogonal. Based on this property, we propose a new speaker normalization method which collapses the subspace that encodes speaker information, without requiring transcriptions. Probing experiments show that our method effectively eliminates speaker information and outperforms a previous baseline in phone discrimination tasks. Moreover, the approach generalizes and can be used to remove information of unseen speakers.