Algorithms with low computational cost for monitoring and analysis of Colombia soundscapes

Studies focused on soundscape are important on biological conservation, because natural sounds are permanent and with dynamic properties, they have been linked to the welfare of the environment and the structure of the landscape. These studies usually analyze the sound in time and frequency domains, with computationally heavy and centralized algorithms. However, new technologies for real time analysis requires distributed algorithms with low computational cost. Hence, the present work evaluates the computational cost of alternative methods with potential applicability in analysis of time-varying signals. The analyzed methods are short time Fourier transform, harmonic expansion, wavelet transform (analytical and non-analytical Morlet, Mexican hat, and Paul) and orthogonal polynomial expansion (Legendre, Chebyshev, and Hermite). A comparison between these methods is presented, in which processing time, memory consumption, quality of reconstruction and grouping index are some of the features selected, resulting in a useful computational cost ranking. The methods are applied to several signals generated with different procedures, such as artificial modulated signals and natural recorded sounds (provided by The Alexander Von Humboldt Institute). In conclusion, Harmonic expansion, Chebyshev expansion, Legendre expansion and Short Time Fourier Transform are the best methods with excellent performance in all features.