Deep Neural Convolutive Matrix Factorization for Articulatory Representation Decomposition

TL;DR

This paper introduces a neural convolutive matrix factorization method to decompose articulatory data into interpretable gestures, bridging articulatory phonology and deep learning for more intelligible speech representations.

Contribution

It presents a novel neural sparse matrix factorization approach to extract phonological gestures from articulatory data, enhancing interpretability and phoneme recognition.

Findings

Gestural scores encode phonological information effectively.

The method improves interpretability of speech representations.

Results demonstrate successful decomposition of articulatory signals.

Abstract

Most of the research on data-driven speech representation learning has focused on raw audios in an end-to-end manner, paying little attention to their internal phonological or gestural structure. This work, investigating the speech representations derived from articulatory kinematics signals, uses a neural implementation of convolutive sparse matrix factorization to decompose the articulatory data into interpretable gestures and gestural scores. By applying sparse constraints, the gestural scores leverage the discrete combinatorial properties of phonological gestures. Phoneme recognition experiments were additionally performed to show that gestural scores indeed code phonological information successfully. The proposed work thus makes a bridge between articulatory phonology and deep neural networks to leverage informative, intelligible, interpretable,and efficient speech representations.