Dhaka University Journal of Applied Science & Engineering

This paper is concerned with the comparison between LMS (t,east Mean Squared) and NLMS (Normalized Least Mean Squared) algorithms on noise cancellation problems. Cancellation of noise was attempted on contaminatetl speech segments. Uttered speech signal samples were recorded from an individual in a quiet atmosphere. 'l'he speech sample signals appeared almost noiseless to a listener when played back through a headphone. These speech samples were takcn as noise less in this work. Computer generated white and filtered noise samples were added purposefully rvith the speech segment, As a result contaminated speech segments were formed. The contaminated speech was processed with an adaptive lilter where LMS and NLMS algorithms were used. The effectiveness of the algorithms was tested by comparing Nrr (Noise Reduction Ratio) attained after liltering the contaminated speech segment using both algorithms. Also human subjects listened to the recovered and noisy speech to grade the result. tn this work, Nrr and mean squared errors (MSE) of the algorithms attained in noise cancellation application were taken as the basis of the comparison of LMS and NLMS algorithms. Variation of results with respect to input SNR and filter length were observed in the work. Convergence speeds for both algorithms were also shown in the work

1. ref1