Font Size: 

Human speech activity in the real condition is distorted by background noise, especially wind noise. This noise affects the speech intelligibility and speech quality especially for hearing aids users. The hearing aid is one implementation of signal processing applications. Elimination of noise from the signal is the major task in signal processing applications. One of the algorithms commonly used in hearing aid technology is the adaptive filter. The adaptive filter was used in hearing aid because this algorithm has a relatively light computation compared to other algorithms. The adaptive filter has several algorithms to remove noise from the signal. In this article, Least Mean Square (LMS) algorithm, Normalized Least Mean Square (NLMS), and Recursive Least Square (RLS) are implemented on the noisy speech dataset. The speech dataset was added with seven different wind noises and produced noisy speech. The seven wind noises consist of four wind noise datasets, two wind noises recorded using a GO-Pro camera and smartphone, and one wind noise simulation. The experiment used the TIMIT dataset and three objective evaluation parameters to evaluate the performance of LMS filters. The three parameters are Short-Time Objective Intelligibility (STOI), Perceptual Evaluation of Speech Quality (PESQ), and Log-spectral distance (LSD). The experimental results showed the performance of LMS filters in the various conditions of wind noise. This experiment was conducted to measure how robust the LMS filter is in removing noise from the signal while maintaining speech quality.