Removal of Real World Noise in Speech: Comparision of Various Parameters Using Kalman and H-Infinity Filter Algorithms

Abstract

Background/Objectives: The performance of the speech enhancement techniques is examined by applying them to the speech signals corrupted by real world noise. Methods: An advanced coding Methodology is introduced to reduce the noisy signal using H-infinity filter. Here two compression techniques - Switch Split Vector Quantization (SSVQ) and Multi Stage Vector Quantization (MSVQ) are used. The enhancement techniques like Kalman Filter and Recursive Filter were assessed for similarities and differences with H-Infinity Filter and the outcomes are compared using signal to noise ratio which is likely to be affected on real world noise. Findings: In this paper using kalman filter, recursive filter, h-infinity filter plays vital role for comparing the various parameter characteristics. Kalman filter uses ordered steps that solve a mathematical problem. H-infinity filter differs from the normal changed Kalman filtering as it requires the knowledge of commotion parameters. H-infinity minimizes the estimation errors and thus obtains robustness and obtains the better results in enhancement. Application/Improvements: This study can be helpful for improving the intelligibility in speech signal and the same method can be implemented with different types of hybrid vector quantization techniques and may achieve better SNR and quality of the signal.