A Perceptually-Motivated Approach for Low-Complexity, Real-Time Enhancement of Fullband Speech

TL;DR

This paper introduces PercepNet, a low-complexity, real-time speech enhancement method that leverages perceptual features to improve fullband speech quality with minimal computational resources.

Contribution

The paper presents PercepNet, a novel perceptually-motivated approach that achieves high-quality, real-time speech enhancement at 48 kHz with less than 5% CPU usage.

Findings

High-quality, real-time enhancement of fullband speech at 48 kHz.

Less than 5% CPU core utilization.

Outperforms traditional spectral subtraction methods.

Abstract

Over the past few years, speech enhancement methods based on deep learning have greatly surpassed traditional methods based on spectral subtraction and spectral estimation. Many of these new techniques operate directly in the the short-time Fourier transform (STFT) domain, resulting in a high computational complexity. In this work, we propose PercepNet, an efficient approach that relies on human perception of speech by focusing on the spectral envelope and on the periodicity of the speech. We demonstrate high-quality, real-time enhancement of fullband (48 kHz) speech with less than 5% of a CPU core.