Synthetic Wave-Geometric Impulse Responses for Improved Speech Dereverberation

TL;DR

This paper introduces a hybrid synthetic dataset for speech dereverberation, combining wave-based and geometric methods to better simulate room acoustics, leading to improved dereverberation performance.

Contribution

The paper proposes a novel hybrid synthetic RIR dataset that enhances speech dereverberation models by accurately simulating low-frequency components.

Findings

Models trained on hybrid synthetic RIRs outperform those trained on purely geometric RIRs.

Accurate low-frequency simulation is crucial for effective dereverberation.

Hybrid dataset improves performance across multiple real-world RIR datasets.

Abstract

We present a novel approach to improve the performance of learning-based speech dereverberation using accurate synthetic datasets. Our approach is designed to recover the reverb-free signal from a reverberant speech signal. We show that accurately simulating the low-frequency components of Room Impulse Responses (RIRs) is important to achieving good dereverberation. We use the GWA dataset that consists of synthetic RIRs generated in a hybrid fashion: an accurate wave-based solver is used to simulate the lower frequencies and geometric ray tracing methods simulate the higher frequencies. We demonstrate that speech dereverberation models trained on hybrid synthetic RIRs outperform models trained on RIRs generated by prior geometric ray tracing methods on four real-world RIR datasets.