2024 INTERSPEECH INTERSPEECH 2024

Spatial Acoustic Enhancement Using Unbiased Relative Harmonic Coefficients

Abstract

This paper targets at enhancing the noisy soundfield over the entire recording area and all the individual channels, while preserving the spatial clues of the original soundfield. For the goal, we utilize a recently proposed spherical harmonics (SH) domain feature denoted relative harmonic coefficients (RHC) as it compactly contains the source’s spatial information. Specifically, we (i) propose an unbiased estimator of RHC in noisy environments; (ii) estimate the source signal in noisy environments using a SH domain beamformer; (iii) enhance the SH coefficients by multiplying the estimated RHC and source signal; and (iv) reconstruct the entire soundfield based on the enhanced SH coefficients. Finally, we evaluate and validate the performance of the enhancement algorithm using extensive simulations.

🧭 Keyword Pioneer — relative harmonic coefficient
🐝 Cross-Pollinator — Artificial Intelligence, Computer Vision, Deep Learning, Machine Learning, Mathematics & Optimization, Natural Language Processing, Speech & Audio