A fundamental frequency detector of speech signals based on short time Fourier transform

The design of an accurate speech fundamental frequency detector is one of the most challenging areas of speech signal processing research. A new fundamental frequency determination method for speech signals is presented in this paper. This new method is based on a wavelet transform followed by a short-time Fourier transform. The wavelet transform is used to eliminate and attenuate the harmonics of the speech signals. The fundamental frequency of the speech signals is determined by estimating the peak amplitude of the short-time spectrum of the wavelet-transformed speech signals. The new method is not only effective for accurate fundamental frequency estimation but it also does not have the problem of occasional incorrect frequency estimation, such as frequency doubling or halving, usually associated with the conventional methods