Speech Enhancement Using Rasta And LPC

Abstract

The enhancement of speech is to recover and improve the speech quality and its intelligibility by using different techniques and algorithms. An iterative Kalman filter model is using the linear predation coefficients (LPC) parameter to estimate the noise present in degraded speech. But LPC parameters are sensitive with respect to different types of noise. To overcome this problem Kalman filter with overlapping frames are used. The cochlear implant is used for those who are suffering from human ear problem and it needs speech enhancement. The main challenge in enhancement of speech is analysis of speech signal features and designing of efficient filters. The complex speech spectrum is used to enhance the noisy speech. At low-frequency analysis of speech, substantially in order to bubble noise and car interior noise so that it contain a lot of energy at low frequency, more accurate result is given by iterative Wiener filter. A new method to change the magnitude and phase, a modified complex spectrum is to provide a better estimation. Another algorithm using feedback particle filter will reduce minimum mean square error (MMSE) using iterative method. A new algorithm is also there in literature using zero replacement signal of noisy speech. This method is using Fourier transform (FT) of cross-correlation function between noisy speech signal and zero replacement signals. The speech enhancement is using a pre-filter method to obtain a better auto regression (AR) co-efficient by using temporal and the simultaneous masking threshold. For the estimation of single channel speech enhancement algorithms, whose parameters are estimated by codebook method, the use of short time Fourier transforms (STFT) to reconstruct the noise observation and fundamental frequency of speech has been done. A non-zero mean condition that builds up by the Bayesian short time spectrum amplitude (STSA) algorithm is introduced in the stochastic deterministic (SD) speech. An analytical function is used for the magnitude of STFT of speech signal in the form of MMSE and phase in maximum-likelihood. Keywords Speech Enhancement; Amplitude and phase estimation; Gaussian process; Stochastic model; Kalman filter; Wiener filter; Particle filter; Cochlear implant.