SonicSieve: Bringing Directional Speech Extraction to Smartphones Using Acoustic Microstructures

Researchers have developed SonicSieve, a passive acoustic microstructure that attaches to standard wired earphones to give smartphones powerful directional speech extraction. The system focuses on a 30° angular region, achieving a 5.0 dB signal quality improvement. Remarkably, its performance using only two microphones exceeds that of conventional five-microphone arrays. An end-to-end neural network processes the raw audio in real-time directly on the mobile device, enabling clear recordings in noisy environments like restaurants.