Real-time Multi Source Speech Enhancement for Voice Personal Assistant by using Linear Array Microphone based on Spatial Signal Processing

摘要

This paper describes an algorithm for suppressing composite noise in multi-source acoustic signals for real-time voice personal assistant application. This paper gives the implementation and algorithm to get desired noise free signal to speech enabled devices by using ICS 52000 linear microphone array and Pico IMX Board, it first finds the direction of arrival of the sources using time delay between the microphone signals, based on the direction angle, the microphone signals are Presteering. The presteered signal is time shifted and aligns same signal is given to noise signal extraction unit by using blocking coefficients, both results are adaptively filtered using NLMS filter, it gives the noise free desired signal. The resultant is given to wake word engine and alexa (cloud), the proposed speech enhancement algorithm shows significant performance improvements in terms of speech processing evaluation and can also be used to deactivate certain processes during the non-speech signal. We tested the output to voice enabled devices the accuracy of detection rate matches the reference devices like alexa echo and ok google.