Abstract
The front-end speech enhancement system is regarded as an essential component for maximizing the performances of smart technology for voice interaction in complicated live acoustic scenes. Existing research has had the following limitations: short-distance detection, poor suppression of non-stationary interferers, and inaccurate estimation of the direction of arrival. To tackle these issues, this paper proposes a 48-channel acoustic array system for directional sound-capture (DSC). This system implements a field programmable gate array (FPGA)-based acquisition and signal processing algorithm: broadband acoustic beamformer based on audio-visual (A-V). To our knowledge, this is the first time that a DSC system that uses A-V for terminal voice interaction has been implemented by FPGA. Experiments were set up in diverse acoustic scenes to evaluate the system’s performance. The results imply that the proposed system can be widely applied to smart scenes in complicated acoustic environments contaminated with intense background noise and competing non-stationary interferers, as well as provide real-time speech recognition and classification.

This publication has 30 references indexed in Scilit: