NOISE CANCELLATION USING LEAST MEAN SQUARE AND WAVELET TRANSFORM FOR SPEECH ENHANCEMENT

Abstract

 This paper presents about Adaptive Filter Algorithms used in Embedded Signal Processing for Speech Enhancement. Filters are generally used to select or to remove or to separate out particular fixed frequency, but in Adaptive Filters the frequency selection is important as well as the coefficients of Adaptive filter are being updated by the Adaptive Algorithms. Adaptive Filters are the filters whose filter coefficients are updated automatically by the process of steepest descent algorithm. An Adaptive Filter is defined as a self- adjusting system that relies for its operation on a recursive algorithm, which makes it possible for the Filter to perform satisfactorily in an environment where knowledge of the relevant statistics is not available. Least Mean Square (LMS) is the algorithm used to update filter coefficients by subtracting the desired signal from input signal producing error signal which updates the algorithm variables at each iteration repeated iterating process trains itself to the input signal and cancels noise. Wavelet transform is taking the overlapped windowed frames of input signal transforming it from time domain to frequency to understand the spectrogram of signal apply thresholding depending upon the parameters to consider and denoise the signal. Databases of clean speech and Noise speech can be downloaded freely from TIMIT, NOIZEUS, and SpEAR database. Implement the both the filters LMS and Wavelet and compare them to conclude which algorithm works well.