Multi-Axis Speech Similarity via Factor-Partitioned Embeddings

TL;DR

This paper introduces a factor-partitioned embedding framework for speech that separates multiple attributes into distinct subspaces, enabling attribute-conditioned retrieval and suppression of biases.

Contribution

It proposes a novel multi-axis embedding method that disentangles speech attributes into subspaces, improving retrieval and bias control over conventional single-vector embeddings.

Findings

Embeddings support attribute-conditioned retrieval with attribute suppression.

Signed axis weighting reduces same-speaker bias in cross-corpus retrieval.

Code implementation is publicly available at the provided GitHub URL.

Abstract

Speech encodes multiple simultaneous attributes -- linguistic content, speaker identity, dialect, gender --that conventional single-vector embeddings conflate. We present a factor-partitioned embedding framework that maps each utterance into a single vector whose subspaces correspond to distinct axes of variation. A shared acoustic encoder feeds per-axis linear projection heads, each trained via distillation from a specialist teacher or a contrastive objective over shared-label pairs. The resulting embeddings support attribute-conditioned retrieval: similarity is computed as a signed weighted sum over per-axis cosine scores, allowing retrieval that jointly considers what was said and how -- or explicitly suppresses one attribute to surface another. We evaluate on cross-corpus retrieval over corpora sharing the Harvard sentence prompts, demonstrating that signed axis weighting can suppress same-speaker bias and surface semantically matched utterances across recording conditions. Code is available at: https://github.com/jimregan/spoken-sentence-transformers