Performance Analysis of Speech Enhancement Techniques

Noise presents a significant challenge in signal processing, especially with speech signals that are prone to various types of acoustical disturbances like babble noise or street noise. The ability to predict and remove noise effectively depends on how rapidly the noise changes relative to the speech. Filtering techniques have been developed to address these challenges and improve speech quality. This paper examines and contrasts three prevalent filtering techniques: The Wiener filter, Kalman filter, and Least Mean Square (LMS) algorithm. It assesses their performance based on various metrics including spectrogram quality, Mean Squared Error (MSE), computation time and Perceptual Evaluation of Speech Quality (PESQ) scores. By conducting simulations with varied noisy speech signals across different noise environments, the research assesses the effectiveness of each approach concerning spectral accuracy, noise attenuation capability, perceived speech quality, and computational efficiency. This evaluation seeks to assist in identifying appropriate speech enhancement methods tailored to particular application requirements and noise profiles.