An important application for use with multimedia databases is a
browsing aid, which allows a user to quickly and efficiently preview
selections from either a database or from the results of a
database query. Methods for facilitating browsing, though, are
necessarily media dependent. We present one such method that
produces short, representative samples (or “audio thumbnails”) of
selections of popular music. This method attempts to identify the
chorus or refrain of a song by identifying repeated sections of the
audio waveform. A reduced spectral representation of the selection
based on a chroma transformation of the spectrum is used to
find repeating patterns. This representation encodes harmonic relationships in a signal, and thus is ideal for popular music which
is often characterized by prominent harmonic progressions. The
method is evaluated over a database of popular music and found to
perform well, with most of the errors resulting from songs that do
not meet our structural assumptions.
