Abstract
In this thesis, an online radio service called virtual channel is proposed. This service provides a continuous musical environment where the playbill is made according to the atmosphere or the user’s habit. To achieve this goal, we first extract and analyze the audio features of multiple data streams from the online radio servers. Through this step, we classify all radio channels and determine whether the content contains musical intension. After that, we continue to analyze the remaining channels which are playing music and detect the musical mood of the music slices by music psychological theory and heuristic rules. The playbill according to the user’s preference can then be arranged. Due to the constraint of the processing time in the streaming environment, we have to process all schemes mentioned above in real time. Therefore, we have to shorten the judging time by extracting few features and using efficient classifiers to satisfy the limitation. Finally, we perform a series of experiments to evaluate the performance of the proposed framework, and the discrimination rate of the schemes. The results show that the mechanism of virtual channel is workable in the streaming environment.