Single-source zones detection using relative harmonic coefficients under diverse acoustic scenarios

Abstract

Accurate detection of single-source zones, where only a single source is active or dominant within the mixed recordings, benefits many acoustic signal processing techniques, while with very little research focusing on this topic. This article uses a higher-order sensor array and proposes a spherical harmonics (SH) domain detector that is applicable under diverse acoustic scenarios. The detector utilizes multi-dimensional scaling (MDS) by projecting a recently introduced SH-domain feature denoted relative harmonic coefficients (RHC) into a low-dimensional space, where the single-source zones can be easily recognized as they locate at distinctive positions compared to the overlapped and source-absent zones. The method eases limitations suffered by the baseline detectors, such as direct-path wave propagation and far-field requirement, and is additionally capable of counting the unknown source number and automatically clustering the detected single-source zones into each source they originally belong. Extensive experimental results in both simulated and real-life environments have confirmed the proposed detector outperforms the baselines, especially in strongly noisy, reverberant and near-field soundfield.