Space-domain cross-pattern coherence post-filter for speech enhancement with linear microphone arrays

Abstract

This paper evaluates the space-domain cross-pattern coherence (SD-CroPaC) algorithm in speech enhancement applications employing linear microphone arrays. The algorithm computes the normalized cross-spectral density between beamformed signals in the time-frequency domain, which originate from non-overlapping sub-arrays. The resulting post-filter is then applied to the output of another beamformer to suppress interference and background noise. The evaluation was performed with a listening test comprising simulated multi-speaker scenarios. The results clearly show the improved performance achieved by the post-filter compared to a conventional baseline beamformer. Additionally, the listening test results suggest that fine-tuning the filter’s parameters may offer a further performance improvement.