Multiple Lyrics Alignment: Automatic Retrieval of Song Lyrics
Sprache des Vortragstitels:
Englisch
Original Tagungtitel:
6th International Conference on Music Information Retrieval (ISMIR 2005), London, UK
Sprache des Tagungstitel:
Englisch
Original Kurzfassung:
We present an approach to automatically retrieve and extract
lyrics of arbitrary songs from the Internet. It is intended
to provide easy and convenient access to lyrics
for users, as well as a basis for further research based
on lyrics, e.g. semantic analysis. Due to the fact that
many lyrics found on the web suffer from individual errors
like typos, we make use of multiple versions from
different sources to eliminate mistakes. This is accomplished
by Multiple Sequence Alignment. The different
sites are aligned and examined for matching sequences of
words, finding those parts on the pages that are likely to
contain the lyrics. This provides a means to find the most
probable version of lyrics, i.e. a version with highest consensus
among different sources. Furthermore, we propose
a measure to estimate the quality of the predicted text.
Forschungs-