12 May 2014
Time: 3:30 - 4:30pm
Venue: Eng. 2.07 Engineering Building, Queen Mary University of London, Mile End Road, London, E1 4NS
Matt McVicar (AIST Japan)
Towards the automatic transcription of lyrics from audio
Transcribing lyrics from musical audio is a challenging research problem which holds promise for the attractive task of lyric-based retrieval. However, automatic lyric transcription has not yet benefited from advances in the related field of automatic speech recognition, owing to the prevalent musical accompaniment and differences between the spoken and sung voice. In this talk I will outline our recent progress in automatic lyric transcription, which exploits repetitive segments for improved accuracy on acapella vocals. I will then outline our current direction of research and finally open the floor for questions and discussion.
Matt received a Master of science in Mathematics in 2008 and Master of Research in Complexity Sciences in 2009, both from the University of Bristol. Just prior to the completion of his PhD at the same institution, he was awarded a scholarship by the USUK Fulbright commission to conduct research at Columbia University in the city of New York. His research interests are focused on data-driven applications to Music Information Retrieval tasks, and in particular the use of freely-available internet resources to solve these problems.