Özet:

Abstract Pitch period evaluation of speech signal is used in many important applications of speech technology. However, among the existing methods only some can work in case of non-linear and non-stationary signals. The main reason is that the pitch detection methods are based on the assumption that speech production process is linear. Selection of pitch period estimation algorithm is always focuses on finding a compromise between time and frequency resolution, robustness, computational complexity and time delay. The aim of this paper is to develop a new method for estimating the pitch period based on empirical wavelet transformation. Method of constructing a family of adapted wavelets assumes that the filters depend on the information location in speech spectrum of the analyzed signal. Empirical wavelets are defined as bandpass filters for each segment of the speech signal. Instantaneous frequency characteristics are considered as pitch period detection features. Teager-Kaiser energy separation operator is used for its extraction. The comparison of this method with other pitch estimation algorithms is presented. References A comparative performance study of several pitch detection algorithms / [L. Rabiner, M. J. Cheng, A. E. Rosenberg, C. A. McGonegal] // IEEE Transactions on Acoustics, Speech and Signal Processing. – 1976. – № 5. – P. 399–417. DOI: 10.1109/TASSP.1976.1162846

Anahtar Kelimeler: