Identification Musicale à l'aide de Technologies Vocales
msra(2009)
摘要
Music identification is a matching process of an song snippet to a known song by the system. There are various use cases of a such system, as protection of copyright, or simply, allow a user to identify the piece of music he listens, and to retrieve information like the artist name or the album title. Because of the interest of a such an application, several approaches have already been studied, most commonly based on methods of pattern recognition. We propose to use speech processing techniques, effective in difficult environments. Our approach is based on Gaussian mixtures (GMM) and hidden Markov models (HMM), which are widely used concepts in the speech processing domain. We applied to the music a speakers type segmentation method. Our experiments demonstrate an identification accuracy of 100% with 25 dB of additive noise and with low bitrate encoded MP3, sugesting our system is resistant to noise and signal compression.
更多查看译文
关键词
indexation,récupération d'information en fonction du contenu,acoustic,parole. keywords: music identification,content-based information retrieval,mots-clés : identification musicale,speech.,acoustique,in- dexation
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络