Skip to Main Content
This paper presents an improvement to ordinary voice filters that are unable to remove low frequency noise. This problem occurs because their filter parameters are over tuned, thereby destroying the speech characteristics. The proposed scheme combines NLMS, Fuzzy logic and Kalman filtering to restrain the background noise and keep the speech characteristics. This scheme is called the Normalized Fuzzy Logic Kalman Filter (NFLKF). It is especially effective when speech signals are collected in a noisy environment. Here, the output signal of the Kalman filtering is analyzed with the normalized LMS to obtain the coefficient, sigmak, and is also analyzed with the fuzzy logic to obtain the threshold, fk. Then, sigmak and fk are used together to adjust the Kalman filter parameters. This scheme can restrain the noise and improve the signal-to-noise ratio. The empirical validation is done by comparing the spectrogram from our scheme with the spectrograms from other filtering schemes, including the Normalized Least Mean Square (NLMS) Filter, the Kalman Filter and the Recursive Least Square Filter (RLS). The results show that the filtering schemes proposed can indeed restrain medium and low frequency noises which are usually difficult to handle, and does not compromise the speech characteristic. Therefore, a better signal-noise ratio is obtained and the speech quality is enhanced.