A Hybrid DSP/Deep Learning Approach to Real-Time Full-Band Speech Enhancement

TL;DR

This paper presents a hybrid DSP and deep learning method for real-time full-band speech enhancement that improves noise suppression quality while maintaining low computational complexity suitable for low-power devices.

Contribution

It introduces a novel hybrid approach combining neural network-based gain estimation with traditional pitch filtering for effective real-time speech enhancement.

Findings

Achieves higher speech quality than traditional spectral estimators.

Operates in real-time at 48 kHz on low-power processors.

Maintains low complexity suitable for practical applications.

Abstract

Despite noise suppression being a mature area in signal processing, it remains highly dependent on fine tuning of estimator algorithms and parameters. In this paper, we demonstrate a hybrid DSP/deep learning approach to noise suppression. A deep neural network with four hidden layers is used to estimate ideal critical band gains, while a more traditional pitch filter attenuates noise between pitch harmonics. The approach achieves significantly higher quality than a traditional minimum mean squared error spectral estimator, while keeping the complexity low enough for real-time operation at 48 kHz on a low-power processor.