Popular Music Retrieval by Detecting Mood

摘要

In the community of music information retrieval, researchers developed methods to retrieval music with a particular melody, they also developed methods to retrieval music by similarity. We aim to retrieval music by mood, which is sometimes the exclusive manner people select music to enjoy, for example, when someone is sad for some reason, she or he wants to listen to a piece of music that can cheer her or him up, at this moment she or he will search music segment by mood no matter what the melody sounds and which the piece of music is similar to. In the essence, the difficulty for music retrieval by mood stems from the gap between the rich meanings that users want when they query and the shallowness of content descriptor we can actually compute. Recently, a new approach, Computational Media Aesthetics (CMA) is presented aiming to fill this gap, the core trait of it is that data is interpreted with its maker's eye [1]. Composers choreograph the expectation to arise emotion, and performers convert the musical intention into music language to arise emotion, which inspires us to analysis music mood on the viewpoint of how music is made, i.e. the viewpoint of CMA avoiding coping with ambiguity of emotion audience arise when confronted with music. Juslin found that two music dimensions could explain the transfer of emotional content from performer to audience: tempo and articulation, tempos were either fast or slow while articulations were either staccato or legato [2], Table 1 depicts the relationship of music mood with tempo and articulation.