Parameter Estimation For Non-stationary Speech Feature

With the rapid development of modern signal processing theory, the detection and parameter estimation for non-stationary signal have a wide range of applications in some engineering field such as sonar, communication, radar, system identification, fault diagnosis, speech recognition etc, and it has got the deep concern of many experts and scholars home and abroad. The detection and parameter estimation for dynamic signal is an important content of modern signal processing. At present, the commonly used methods for analysis the signals in time domain and frequency domain have some limitations. Some methods such as short-time windowing can not detect the signal and estimate the corresponding parameter effectively, which smooths the non-stationary signal. Fractional Fourier transform is particularly suitable for chirp type signal processing, because fractional Fourier transform has some property that traditional Fourier transform do not has, it also can be understood as the chirp-based decomposition.The signals of human or animal sound can be modeled as the fundamental and its harmonic components, and the frequency is time-varying, so the traditional Fourier transform can not be a good description of sound signals. Fractional Fourier transform and its related algorithm can be used effectively for speech signal detection and parameter estimation, because fractional Fourier transform satisfy the linear conditions and suitable for dealing with time-varying signals, in addition, it can avoid the interference of cross terms in the case of noise and multicomponent. In this paper, we estimated the parameters of different speaker's voice by fractional Fourier transform and its fractional autocorrelation algorithm according to the time-varying characteristic in speech signal, and the test statistic of the instaneous peak was given. Simulation by matlab, we observed the change of the frequency. The results show that using fractional autocorrelation algrithm in the speech signal detection and parameter estimation has its unique advantages, and this method is effective and possible.The research work in this paper can not only be directly applied to speech signals and other sound signals detection, but also has a certain theoretical and practical significance for the realization of future work in variety signal detection and processing. It is a important means in developing the information technology, searching and exploring new scientific and technological achievements in the future. It has far-reaching and important significance for promoting the development of the entire signal related fields.