Music Content Analysis on Audio Quality and Its Application to Music Retrieval

Abstract

Nowadays, more and more users are uploading their music recordings of live music concerts to video sharing websites such as YouTube. The audio quality of these uploads, however, varies widely due to their recording conditions, and most existing video search engines do not take the audio quality into consideration when ranking their search results. We proposed a learning-to-rank technique for ranking live music recordings on YouTube according to audio quality and constructed three datasets. Several ?song-level? audio feature representations and learning-to-rank algorithms were studied. Specifically, we explore various segmentation methods and fusion system to account for the temporal characteristics of live music. A comprehensive performance study is conducted to demonstrate the effectiveness of the proposed system through both objective and subjective evaluations. Moreover, we apply the result into our practical system i2MUSE to help the users to search and find good music to do exercise.