Classification of Public Radio Broadcast Context for Onset Detection

Abstract

Abstract The rapid development of the modern information and communication technologies has in uenced the various aspects of the human communication and behavior, including the mass media communication and journalism. This development allowed the deployment of di erent mass communication applications by motivating people in the content analysis of di erent communication media. Radio broadcasting can be identi ed as a communication media which is so close to every citizen in a country. Content analysis in the radio broadcast context for various commercial application development (i.e. news monitoring, song monitoring, speaker recognition etc.) emerged as a major research area which facilitates the broadcast monitoring process. This dissertation focuses on the investigation of a uni ed methodology for the onset detection in Sri Lankan radio broadcast context with the approach of classi cation of the broadcast context. Various audio patterns in the broadcast context were observed and a supervised learning approach was employed in the classi cation process. Di erent audio features were examined with respect to the broadcast context. Identi ed audio semantics in the broadcast context were used in re ning the output gained in supervised learning models. Onsets were predicted using the classi cation results. The evaluation method was carried out with ground truth data obtained from a Sri Lankan FM broadcast recording. The proposed approach provided the accuracies of 41% for news, 76% for radio commercials, 75% for songs and 59% for other voice related segment classi cation. The onset detection model was successful in predicting the onsets with an error rate of (+/-) 2.5s with approximately 82% of accuracy level. The proposed strategy can be easily adapted in broader audio detection and classi cation tasks including additional real world speech-communication scenarios with some improvements to the proposed classi cation model.