Sparse Mixture of Local Experts for Efficient Speech Enhancement

TL;DR

This paper presents a sparse mixture of local expert neural networks for speech denoising, which improves performance and reduces complexity by specialized sub-models guided by a gating network based on speech degradation or speaker gender.

Contribution

The paper introduces a novel ensemble model with a gating network that assigns speech signals to specialized neural networks, enhancing denoising performance with fewer parameters.

Findings

Ensemble of specialist networks outperforms a generalist network in speech denoising.

The proposed model reduces computational complexity while maintaining high denoising quality.

Gating network effectively classifies subproblems based on speech degradation or speaker gender.

Abstract

In this paper, we investigate a deep learning approach for speech denoising through an efficient ensemble of specialist neural networks. By splitting up the speech denoising task into non-overlapping subproblems and introducing a classifier, we are able to improve denoising performance while also reducing computational complexity. More specifically, the proposed model incorporates a gating network which assigns noisy speech signals to an appropriate specialist network based on either speech degradation level or speaker gender. In our experiments, a baseline recurrent network is compared against an ensemble of similarly-designed smaller recurrent networks regulated by the auxiliary gating network. Using stochastically generated batches from a large noisy speech corpus, the proposed model learns to estimate a time-frequency masking matrix based on the magnitude spectrogram of an input mixture signal. Both baseline and specialist networks are trained to estimate the ideal ratio mask, while the gating network is trained to perform subproblem classification. Our findings demonstrate that a fine-tuned ensemble network is able to exceed the speech denoising capabilities of a generalist network, doing so with fewer model parameters.