CLASIFICADOR DIFUSO DE SEÑALES ACÚSTICAS AMBIENTALES BASADO EN ANÁLISIS DE COMPONENTES INDEPENDIENTES

Abstract

The noise evaluation is usually done by considering the impact of a specific noise source. However, in almost all environments, a large number of different sources contribute to noise in a particular area. The main causes of noise pollution are those related to human activities such as transport, trade, leisure activities, among others. Because of this, there is the need to obtain the source separation of sources that appear mixed in the atmosphere, as well as knowing the type of source. This information is useful to take the necessary actions to help reduce noise levels. This work shows a method for blind source separation of sources in field, in a controlled environment. It is mainly divided into two stages, the first is a pre-processing consists of filtering, wavelets and deconvolution; the second stage uses Independent Component Analysis (ICA). The experiments were performed using five types of sources common in urban areas like: motor vehicles, horns, whistles, sirens and voice. Separation was obtained with 2, 3 and 4 sources mixed in controlled environments, recorded by industrial-type microphone with directional behavior. Besides, a classifier was designed based on fuzzy clustering techniques, able to identify five types of sources considered, with a good performance. The extraction characteristics of the sources is a fundamental part of the classifier performance, this work shows the results obtained using different features. The results presented are satisfactory in the separation of mixtures in field and in their classification. They are an approach to help solve the problem of noise in real environments