Self-supervised Neural Factor Analysis for Disentangling Utterance-level Speech Representations

TL;DR

This paper introduces a novel self-supervised factor analysis approach that disentangles utterance-level speech representations, enabling improved performance on speaker, emotion, and language recognition tasks with limited labeled data.

Contribution

The proposed FA-based model uses hidden acoustic units for utterance-level learning, enhancing SSL speech models for non-semantic tasks without extensive supervision.

Findings

Outperforms WavLM on SUPERB benchmark tasks

Achieves high accuracy with only 20% labeled data

Effectively disentangles speech content from speaker/emotion/language features

Abstract

Self-supervised learning (SSL) speech models such as wav2vec and HuBERT have demonstrated state-of-the-art performance on automatic speech recognition (ASR) and proved to be extremely useful in low label-resource settings. However, the success of SSL models has yet to transfer to utterance-level tasks such as speaker, emotion, and language recognition, which still require supervised fine-tuning of the SSL models to obtain good performance. We argue that the problem is caused by the lack of disentangled representations and an utterance-level learning objective for these tasks. Inspired by how HuBERT uses clustering to discover hidden acoustic units, we formulate a factor analysis (FA) model that uses the discovered hidden acoustic units to align the SSL features. The underlying utterance-level representations are disentangled from the content of speech using probabilistic inference on the aligned features. Furthermore, the variational lower bound derived from the FA model provides an utterance-level objective, allowing error gradients to be backpropagated to the Transformer layers to learn highly discriminative acoustic units. When used in conjunction with HuBERT's masked prediction training, our models outperform the current best model, WavLM, on all utterance-level non-semantic tasks on the SUPERB benchmark with only 20% of labeled data.