Auto-generation of professional background music for home-made videos

作者：Feng Jiashi^*; Ni Bingbing; Van Shuicheng

来源：2nd International Conference on Internet Multimedia Computing and Service, ICIMCS 2010, 2010-12-30 to 2010-12-31.

DOI：10.1145/1937728.1937732

• Audio analysis
• Audio track
• Auto-generation
• Automated systems
• Background music
• Background musics
• Digital video recorders
• Joint probabilistic
• Joint probability
• Online video
• Prior knowledge
• Social media
• Transition modes
• User study
• Video analysis

Techniques to auto-generate professional background music for the home-made video become highly demanded recently with the prevalence of digital video recorders and social media communities. An automated system far such purpose can significantly relieve users'burden of editing background music to accompany home-made video. However, the major obstacle lies in the fact that the assessment of the video-audio matching quality itself is always subjective. We therefore seek solutions from the rich online sharing professional videos to alleviate this difficulty. Mote specifically, a learning-from-Internet framework is proposed to uncover the underlying structures and rules of the video and audio track association patterns from professional online videos. After collecting a large corpus of online professional videos, a joint probabilistic framework is proposed to model prior knowledge from two aspects, namely, the correlation between video and audio tracks, as well as the transition mode from speech and music components. For novel pure-video input, candidate background music tracks are first selected according to the learnt joint probability model, which are further integrated into a smoothed matching sequence via dynamic programming. Qualitative experiments and comprehensive user studies well demonstrate the effectiveness of the proposed framework for background music auto-generation.

• 出版日期2010