SG-VAD: Stochastic Gates Based Speech Activity Detection

TL;DR

This paper introduces SG-VAD, a low-resource speech activity detection model that uses stochastic gates to identify nuisance features, outperforming previous methods with a compact architecture.

Contribution

The paper presents a novel VAD model that models speech detection as a denoising task, with a lightweight design and improved performance.

Findings

Outperforms previous VAD methods on AVA-Speech dataset

Contains only 7.8K parameters, suitable for low-resource environments

Provides comprehensive architecture, experimental results, and ablation studies.

Abstract

We propose a novel voice activity detection (VAD) model in a low-resource environment. Our key idea is to model VAD as a denoising task, and construct a network that is designed to identify nuisance features for a speech classification task. We train the model to simultaneously identify irrelevant features while predicting the type of speech event. Our model contains only 7.8K parameters, outperforms the previously proposed methods on the AVA-Speech evaluation set, and provides comparative results on the HAVIC dataset. We present its architecture, experimental results, and ablation study on the model's components. We publish the code and the models here https://www.github.com/jsvir/vad.