[PD] audio fingerprints

David Sabine dave at davesabine.com
Fri Nov 29 05:30:40 CET 2002


Hello all,

The past discussion on this topic has been primarily about the 'database'
which might store the audio fingerprints.

However, the initial problem is not storage or retrieval of the fingerprint,
but what does the fingerprint LOOK like?  How does one create a 'stamp' or
'summary' or 'fingerprint' of a sound file?  For example, if I were to make
a 'fingerprint' of a 4 minute song...I might map the peak frequencies along
a time code; or perhaps generate a 3 dimensional 'topography' of the entire
song via FFT?  These might be valuable references when comparing multiple
audio 'fingerprints'.   However, it seems like a clumsy way to do it - a
method that might be reliable but would be incredibly slow if one were to
compare one 'fingerprint' to a database of a thousand songs.

Does anybody have any experience with this?

Ideally, an audio 'fingerprint' would be a wonderfully small sequence of
very important information that could be stored in any number of ways -
perhaps a database, perhaps the new [pool] object, perhaps text files.  But
the storage of such a 'fingerprint' doesn't seem to ME to be as mysterious
as the 'fingerprint' itself.  What IS it?  is it a coordinate map?  is it an
array of peak frequencies?

Regards,
Dave S






More information about the Pd-list mailing list