Adaptive noise estimation and voice noise reduction method based on T-S fuzzy neural network

Abstract

The invention provides a noise estimation method based on a T-S fuzzy neural network, and the method comprises the steps: obtaining a time domain noisy voice signal, carrying out the framing of the time domain noisy voice signal, and windowing a current frame; converting a preprocessed time domain noisy voice signal of the current frame into a frequency domain noisy voice signal of the current frame; inputting the frequency domain noisy voice signal into a neural network; calculating a correlation value according to the frequency domain noise estimation signal of the previous frame and the frequency domain noisy voice signal; enabling the neural network to receive the correlation value and correct parameters of the neural network according to the correlation value until the neural networkconverges; and when the neural network converges, calculating a fuzzy weighting coefficient of the current frame noise according to a fuzzy control rule, calculating initial estimated noise of the current frame according to a learning rule, and calculating a frequency domain noise estimation signal of the current frame in combination with the fuzzy weighting coefficient of the current frame noise.Noise estimation is good in real-time performance, high in accuracy, insensitive to a noise environment and good in user experience.

Description

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine

Login to View More

AI Technical Summary

Benefits of technology

This patented technology describes an algorithm for estimating background sound without relying solely upon previous sounds or assumptions about them during training. It uses a special kind of neuron called Fuzzy Neural Network (FNN) which helps with this process. By adjusting certain factors such as gain values and weights associated with each layer within the model, the technique becomes better at identifying different types of ambient acoustic signals while reducing errors caused by reverberations. Overall, this approach provides improved signal quality and reduced error rates compared to existing methods like wavelets and hidden Markov models.

Problems solved by technology

Technological Problem addressed in this patented technical solution describes two main challenges related to improving sound pick up and transmitting performance without reducing background sounds like conversations during phone calls while maintaining good user experience (UX). One challenge involves efficiently estimating noise values from unprocessed signals captured through acoustic sensors placed near loudspeakers. Another issue includes capturing significant changes over time due to environmental factors like temperature variations.

Images