Using fractional Fourier transform in time-frequency analysis for bird songs

This paper presents a spectral-based time-analysis method using fractional Fourier transform (FrFT). This work is inspired by the fact that traditional Fourier transform is not very suitable to deal with non-stationary signals, such as the sound signals, while FrFT has outstanding ability for processing chirp-like signals. The FrFT-based spectrogram is derived and compared with its FT-based counterpart. Experimental results with songs of birds show that the FrFT-based spectrogram with an optimal transform order can achieve higher resolution. Spectrum gathering method is used to show the energy distribution related to the pitch frequency, and the pitch evolving trace is obtained by searching the peak in the gathered spectrum. FrFT has great advantage in detecting and separating chirp signals, which is verified by the gathered spectrum of a mixture of two birdcall signals. The effect of window length on the performance is also discussed.