RARE

Audio fingerprinting (AF) attempts to identify audio clips, either in files or in audio streams. The fingerprints are constructed previously from clean copies of the clips; in this work, fingerprints of length 256 bytes are used, although the length is easily to change. Our audio fingerprinting works with any sort of audio.

AF has many possible applications. Software music players can use AF to identify metadata such as artist, album and track when other means fail (as happens today for user-generated CDs). Stream AF can be used with portable devices such as PDAs to identify broadcast music. Companies can use stream AF to detect whether commercials they have paid for actually air in the market and at the time they expect, and to detect if their commercials have been shortened. AF can be used to completely automate the detection of (noisy) copies of audio in a database, which can be very useful for large databases containing unlabeled data. We've also shown how it can be used to detect choruses in music. Stream AF could also be used to greatly increase the accuracy of the statistical sampling of broadcast music that is currently used to assess royalties for artists.