Toclmiqucs to auto-lCDCfAte professional bacqround. music for 1ht home-made video become highly dCQIPD&-d n:centJ.y with the pKYal.cnce of diptal video ICCOIdcrs and social. media colllIIlWli.1iclJ. An automated syltcm. for such purpose can ⁢nificantly relieve DierS' burden of editing background music to accompany homo-made video.HowcvC'Z, the major obstacle lies in the fact that the assessment of the video-audio mat:chin!; quality itself is always subjective. We thel'Cforc seek IOlutions from the rich onlinc Iharing professional video. to aIlcvimc driB difficulty. More specifically, • Icami.n,&-fmm-Intcmct framc:work is proposed to uncover the underlying stmcturcs and rules of the video and audio b:ack asaociation patterns from professional online videos. After collecting al.arzc c0r-pus of online professional videol, • joint probabilistic framework is proposed to model prior knowlodac from two aspocts, namely. the com:lati.on betwcc:a video IIld ludio tracks, as well AI the trmsition mode from speech and music components. Fa!: novel purevideo input. candidate backgound music tracks .rc first sclcctcd according to the learnt joint probability model, which are further integrated into • smoothed matching scqucnc:e via dynamic prograIIllIliq:. Qu.litative c.xpcrimcnts md comprchcnaive Dim: studies well dcmonsIratc the effectiveness of the proposed framework for background mulic IUto-generation.