A Birdsong Denoising using Cascaded Combination of Phase Spectrum Compensation and Wavelets

Automatic bird call recording and analysis is becoming an effective way of understanding changes in bird population and determining the health of the environment. The behavioral and ecological studies could benefit for identifying species from acoustic recordings. Noise interference that can mask vocalizations of interest is presently troublesome with the automated analysis of bird recordings. Therefore, noise reduction will significantly improve the accuracy of automated analyses and minimize processing work for later phases in bioacoustics. One major obstacle to achieve this noise reduction is the low level of signal-to-noise ratios in unsupervised recordings. In addition, the field recordings of birdsong are also very noisy incorporating ambient noise, activities related to humans and other species, along with noise from the recorder itself. Signal processing is needed to suppress the background noise and enhance the target bird call. So far, some traditional enhancement filters are used. However, their effectiveness is not good for all types of bird spices in bird recordings. We proposed a technique of denoising that used a cascaded combination of wavelet denoising and Phase Spectrum Compensation (PSC). An objective measure of speech quality, time domain and spectrogram analysis using real data clearly, shows that the proposed combination PSC-Wavelet perform well over the individual algorithms resulting in improved quality over natural noisy bird recordings.