Separation and identification of environmental noise signals using independent component analysis and data mining techniques

In the present work, we show a way to separate noise signals recorded with microphones industrial, in order that they can be analyzed separately. Blind Source Separation is accomplished using Independent Component Analysis (ICA) technique in the wavelet domain. Also, it is necessary to identify the separate sources, taking into account that each signal separate has some components of the signals belonging to the initial mixture. Through data mining techniques and characteristic features of the signals obtained are derived rules in order to identify the main source that is present in the mix, for this we propose the use of data mining techniques. The results show a substantial improvement in the separation of mixtures of real environmental noise using ICA, although the mixtures are not fully independent.