Instantaneous Fundamental Frequency Estimation based on FBSE-FAWT

Instantaneous fundamental frequency (IFF) estimation plays an important role in speech processing. This paper proposes a method for estimating a speech signal’s IFF based on the Fourier-Bessel series expansion-based flexible analytic wavelet transform (FBSE-FAWT). The proposed method uses an FBSE- FAWT-based filter bank to decompose the signals into subbands. Components corresponding to the fundamental frequency (FF) can be present in more than one subband. Therefore, a set of subbands are selected whose mean frequency is found to be in the range of 50-500 Hz. In each subband, there may be components separated in the time domain. So for the separation of these components, a new method for time-domain segmentation (TDS) using the Teager-Kaiser energy operator (TKEO) is introduced. Then Hilbert transform separation algorithm (HTSA) based processing is performed on these separated components to obtain the IFF of the speech signal. As a post-processing step, cubic spline interpolation is performed to get accurate IFF. To validate the proposed method, the Keele database is used and the results are compared with the traditional methods. Experimentally it is proved that the IFF estimation method based on FBSE- FAWT is working better and improves the time-resolution of IFF estimation.