Feature Learning with Multi-objective Evolutionary Computation in the generation of Acoustic Features

To choice audio features has been a very interesting theme for audio classification experts. This process is probably the most important to solve the classification problem. In this sense, techniques of Feature Learning generate attributes more appropriate for classification model. Generally these techniques do not depend on knowledge domain and can apply in various types of data. Yet, less agnostic approaches learn a knowledge restricted to the area studded and audio data requires a specific knowledge. Many techniques aim to improve the performance in generation of new acoustic features, among there is the technique based in evolutionary algorithms to explore analytical space of function. Despite the efforts made, there are still opportunities for improvement. This work proposes and evaluates a multi-objective alternative to the exploitation of analytical audio features. Experiments were arranged to validate the method, with the help a computational prototype implementing the proposed solution. Then it was verified the model effectiveness and was shown there is still opportunity for improvement in the chosen segment.