Direct source and early reflections localization using deep deconvolution network under reverberant environment

TL;DR

This paper introduces a deep deconvolution network for localizing the direct sound source and early reflections in reverberant environments, utilizing high-order Ambisonics signals to improve spatial accuracy.

Contribution

The paper presents a novel DCNN model that jointly estimates direct sources and early reflections using HOA covariance matrices, enhancing localization in reverberant spaces.

Findings

The proposed DCNN achieves high accuracy in simulated environments.

The model demonstrates robustness in measured reverberant scenarios.

Localization performance is improved over traditional methods.

Abstract

This paper proposes a deconvolution-based network (DCNN) model for DOA estimation of direct source and early reflections under reverberant scenarios. Considering that the first-order reflections of the sound source also contain spatial directivity like the direct source, we treat both of them as the sources in the learning process. We use the covariance matrix of high order Ambisonics (HOA) signals in the time domain as the input feature of the network, which is concise while containing precise spatial information under reverberant scenarios. Besides, we use the deconvolution-based network for the spatial pseudo-spectrum (SPS) reconstruction in the 2D polar space, based on which the spatial relationship between elevation and azimuth can be depicted. We have carried out a series of experiments based on simulated and measured data under different reverberant scenarios, which prove the robustness and accuracy of the proposed DCNN model.