New AI Tech Can Make Your Calls Sound Crystal Clear

A new paper tackles a major challenge in spatial audio: capturing clear sound in strongly reverberant environments. Neural directional filtering (NDF) can reconstruct a virtual directional microphone with a desired directivity pattern, accurately preserving spatial cues in multi-source scenes — but heavy reverberation makes those cues perceptually hard to distinguish. The proposed neural dereverberation and directional filtering (NDDF) approach reconstructs dereverberated virtual directional microphone signals. Experiments show it consistently beats cascaded dereverberation-and-filtering baselines, and a GAN-based version outperforms the discriminatively trained variant for high-order directional targets. The authors also introduce a method to estimate the directivity pattern using only the input and output signals, which suits signal-mapping-based spatial filtering that synthesizes the output directly without explicit filtering or masking. The paper was accepted to IWAENC 2026.