cs.CLNov 6, 2026

Asymmetric Correspondence Learning for Lexical and Style Embeddings: Interpreting Latent Space via Acoustic Features

Authors: Michael Whealing, Robert Gaizauskas, Thomas Hain

Organizations: School of Computer Science, University of Sheffield

Abstract

Self-supervised speech embeddings jointly encode lexical, speaker, and prosodic information, complicating feature isolation. We introduce Asymmetric Correspondence Learning (ACL), distilling HuBERT features into separate Acoustic Word Embeddings (AWEs) and Acoustic Style Embeddings (ASEs) via a split-latent correspondence bottleneck. Through probing and clustering evaluations on single-word utterances, we demonstrate that substantial variance from extrinsic sources remains encoded within each latent space, despite linear probing indicating the information has been stripped. Whilst the AWE space exhibits cohesive lexical clustering (AP≈\approx0.90, Fisher≈\approx6.28), linear probing on the ASE space achieves 97.9% speaker accuracy across a diffuse topology (Fisher≈\approx1.01), yet preserves substantial lexical variance (Fisher≈\approx0.86) that remains linearly unrecoverable (0.17%). To explain these findings, we apply group-level permutation feature importance over eGeMAPS acoustic features, finding that AWEs partition by vocal tract filters and ASEs by F0F_0 dynamics. These findings suggest that standard diagnostic benchmarks may be insufficient for verifying feature isolation, as substantial speech structure can be retained even when it remains entirely hidden from probes.

Figures & tables

Explore similar work

CardsList
  1. Post-hoc Disentanglement of Textual and Acoustic Features in Pre-trained Speech Encoders

    Oct 3, 2024Hosein Mohebbi, Grzegorz Chrupała, Willem Zuidema +2Representation DisentanglementDisentangled Representation Learning

  2. Probing Low-Level Acoustic Attribute Encoding in CLAP Audio Embeddings

    Jul 4, 2026Héctor Martel, Joe Hennessy-Priest, Taemin ChoRepresentation GeometryAudio Representation Learning