Application of Feed-Forward Neural Network and MMI-Supervised Vector Quantizer to the Task of Content Based Audio Segmentation by Co-operative Unmanned Flying Robots

This paper deals with the preliminary experiments on general audio segmentation using a MMI-supervised tree-based vector quantizer and feed-forward neural network. This method has been tested with the aim of detection of environmental sounds and speech in a sound stream. The segmentation of an audio stream is needed for successful localization of speech or environmental sounds in a stream and their possible future classification or even separation. This method has been developed as a preliminary solution of the task of real-world audio signal segmentation by a set of co-operative unmanned flying robots. Application of the proposed method has been tested in simulating software NESCUAR 1.0. (Natural Environment Simulator for Cooperative Unmanned Aerial Robots, version 1.0), a simulating software tool developed by the authors of this paper. The presented method can be also applied separately; its application is not dependent on the simulating software NESCUAR 1.0.