Application of Fourier-Bessel transform for signal to noise improvement of speech signal

The authors study the representation of the speech waveform using a basis set or kernel function derived from Bessel functions of the first kind. These functions have regular zero-crossings but decaying amplitudes that provide a better match to the behavior of speech waveforms. In addition to exploitation of such improved efficiency for storage and compression systems, the authors have shown the possibility of improving signal-to-noise ratio by filtering which attenuates more of the noise power without attenuating the desired signal. Ideal specifications of signal and additive noise spectra and the ideal amplitude characteristic of a low-pass filter are used to illustrate the possible improvement of signal-to-noise ratio. Then, the improvement in convergence rate of the Fourier-Bessel transform is illustrated by comparing the Fourier and Fourier-Bessel transforms of a set of signals which become more speech-like. Simulation results showing the improvement in a signal-to-noise ratio of speech segments are presented