A method, system, terminal device, and medium for dual-target speech enhancement in smart glasses based on a multi-microphone array.

Abstract

This invention discloses a method, system, terminal device, and medium for dual-target speech enhancement in smart glasses based on a multi-microphone array, relating to the field of speech enhancement technology. The method is applied to smart glasses and includes: acquiring noisy multi-channel raw time-domain speech signals through a multi-microphone array of the smart glasses, obtaining a spectrum set through time-frequency domain transformation; extracting the amplitude and phase features of the complex spectrum of each channel in the frequency domain and introducing local temporal context to construct multi-channel spatial acoustic features with decoupled amplitude and phase; modeling the dual-region complex spectrum using a pre-trained deep neural network to obtain the target complex spectrum, and reconstructing it through inverse time-frequency domain transformation to obtain the enhanced independent wearer speech signal and the target speech signal in a preset direction. This invention achieves parallel dual-target speech enhancement and non-target interference suppression, with low power consumption, high real-time performance, adaptability to the stringent performance constraints of smart glasses, and excellent robustness of speech enhancement in complex noise scenarios.

Images