Expert system for intelligent audio codification based in speech/music discrimination

Automatic Speech/Music discrimination has become a research topic of interest in the last years. This paper present a new approach for speech/music discrimination, which is based on an expert system that incorporates fuzzy rules into its knowledge base. The proposed scheme consists of three stages: 1) features extraction, 2) audio signal classification, and 3) selection of the best audio coder every 23 ms. The fuzzy expert system improves the accuracy rate of a GMM classifier when included into the classification stage. In order to select the best audio coder, the expert system takes information of the current and past frames into account. It is important to emphasize that the low computational cost of the proposed approach makes it feasible for real time applications.