A low-cost automatic switched adaptive filtering technique for denoising impaired speech signals

Speech signals are widely used for several advanced signal processing technologies due to the digital era’s development. In recording speech signals, corruption of the signals by surrounding noise is a severe issue. The signal distorted by noise cannot be examined or used for advanced processing, and these signals must be denoised before being used for further processing. Adaptive noise cancellers are ideal for signal denoising and recovering a corrupted speech signal. This paper introduces an optimal adaptive filter structure using least mean square (LMS) and normalized LMS (NLMS) algorithms to compute the best estimate of the clean signal. A noise-corrupted signal is sent across multiple adaptive filters connected in series. Various stages are added automatically, and the filtering algorithm for each stage is also adjusted automatically. The proposed LMS–NLMS automatic switched adaptive filter model is tested for reducing the noise from a normal speech signal taken from the NOIZEUS database and an abnormal speech signal recorded from Parkinson’s Disease affected patients. Parkinson’s disease (PD) is a severe neurological disorder affecting millions worldwide. Early biomarkers of the disease have been identified to be voice impairments. Hence we have used PD-affected speech signals to prove the effectiveness of the suggested filter model. Both the signals are corrupted by Gaussian signals of different input SNR levels. The simulation results confirm that the proposed filter model performs remarkably well and provides 7–15 dB higher SNR values than the existing cascaded LMS filter models. The MSE value is improved by 30–75%, and the PSNR values are increased by 5 dB. The advantage of using LMS and NLMS adaptive filter in the proposed filter model is that it offers a cost-effective hardware implementation of adaptive noise canceller with high accuracy.