Real Time Speech Signal Denoising Using Wavel...

Real Time Speech Signal Denoising Using Wavelet Thresholding and LMS Adaptive Filtering

Author Name : T. S. R. Krishna Prasad, D. Praneetha, A. Mavya Sri, G. Vanitha, Ch. Jaswanth

DOI: https://doi.org/10.56025/IJARESM.2025.1304250100

ABSTRACT In contemporary communication systems, the clarity of speech signals is paramount, particularly in real-time applications where noise interference can significantly degrade performance [1],[2],[3],[4]. While speech denoising is essential, the dynamic and non-stationary nature of real-world noise presents a substantial challenge. This works presents a real-time speech denoising system that integrates wavelet thresholding with adaptive LMS filtering [8],[10],[6]. Leveraging the multi-resolution capabilities of wavelet transforms, an enhanced thresholding method, incorporating noise estimation and logarithmic scaling [7],[3], effectively reduces noise while preserving critical speech components. Subsequently, an adaptive Least Mean Squares (LMS) filter refines the signal further, dynamically adjusting to residual noise. The system demonstrates significant improvements in Signal-to-Noise Ratio (SNR), Mean Squared Error (MSE), Noise Reduction Ratio (NRR) and Log Spectral Distance (LSD) indicating effective real-time processing [4],[11],[6]. The combined wavelet and LMS approach achieves a balance between noise suppression and the preservation of natural speech characteristics, making it suitable for applications such as teleconferencing and assistive listening devices [9],[13].