De-noising of impaired Speech Signal using Optimized Adaptive Filtering Configuration

Influences of Parkinson's disease (PD) on a patient can mainly be distinguishable in the patient's nervous system, and signs and symptoms can vary from patient to patient. The earliest symptom is that the patient's speech is monotonous; the patient may speak softly, quickly and slur or hesitate before speaking. While a PD patient's speech is being recorded, there are background noises that makes it difficult to interpret the patient's dialogue. The goal is to denoise these distorted speech signals using adaptive filtering techniques. One of the adaptive filtering techniques is the variable-stage-variable-step-size Least Mean Square (VSS-LMS) algorithm, which is the most suitable for denoising of impaired speech signal among other existing speech signal augmentation approaches. Due to its minimal mathematical intricacy and robustness, the VSS-LMS filter is favored. This filter is qualified to achieve enhanced rate of pace in convergence and lesser MSE (Mean Square Error). This proposition depicts an innovative adaptive filtering configuration for de-noising the deteriorated speech signals; which is a multi-stage Variable-Step Size LMS algorithm that has the objective to estimate the signal with lessened noise at a rapid pace. The filter prototype put forth is experimented on a speech signal specimen derived from the various databases. Numerous multiple sources of background noise impair the signals. The suggested filter's efficacy at signal de-noising is judged through the parameters like MSE, SNR, PSNR, MAE, and correlation coefficient. The outcomes confirmed that the suggested structure approach surpasses other conventional adaptive filtering architectures in aspects of output and effectiveness. Consequently, the presented filter design requires use of the VSS-LMS algorithm as a technique that offers a suitable method for quick and affordable de-noising the damaged speech signals.