Thesis Details

In urban soundscapes, there are immersed noise from different sources mainly generated by vehicles, whistles, sirens, car horns and crowds. Nowadays, the environmental noise analysis is based on recordings made by monitoring systems, where different sources are mixed. This condition makes difficult to analyze the predominant sources of noise in an individual mode, which is useful to take decisions for reduce and control the environmental noise. Thus, it helps to improve the health and well-being of the population. This thesis has the objective of separate and classifies noise sources from mixtures recorded in semi controlled environments in no ideal conditions.

In order to separate sources, a method based on blind deconvolution and blind source separation in the wavelet domain is proposed. Also, this method can be applied in real environments under specific conditions. Besides, a method based on Orthogonal Matching Pursuit (OMP) using specialized dictionaries to classify predominant sources, is proposed. The validation of the proposed method is shown by experiments applied to real signals in semi controlled environments and historical information about predominant sources in urban soundscapes.

The proposed method has shown better performance in separation and classification of predominant urban sources. With this approach, it is providing basis for improving the monitoring and analysis results of common noise sources in urban areas