AI-powered Beamforming for Listening to Moving Talkers

Speech enhancement extracts the voice we want to listen to from the background noise and is essential for speech applications, such as automatic speech recognition, to work effectively in noisy environments.This article introduces a novel beamforming technique that tracks the speaker's movement and keeps extracting the target speaker's voice, even when the speaker is moving while talking.Beamforming requires spatial information of the target source and interfering signals such as the direction of arrival.We discuss our previously proposed method to estimate time-varying spatial information by incorporating powerful artificial intelligence technology.This method enables high-performance beamforming even when the target speaker is moving.