Slashdot Mirror
Audio Fingerprinting Via Cell Phones
aruil writes: "MSNBC has a story reporting on yet another audio fingerprinting application. Next year, Royal Philips Electronics will begin selling licenses to allow users to identify songs using their cell phones. Similar technology has already been open-sourced in FreeAmp, which uses the Relatable engine."
no watermark/fingerprint/anything in the music - just an algorithm that samples it, creates a small chunk of data describing it and looks that up against a database for matches. So no loss of quality involved, and if the algorithms as good as they say it'll cope with quite badly broken up music (AM radio? Noisy club?) as input.
Bitzi (the "another" link in the article's "yet another" statement) isn't primarily an audio fingerprinting application. It's a file metadata catalog, audio fingerprints being just one sort of metadata collected. File metadata is keyed by a "bitprint" composed of two cryptographic hashes. The code for generating bitprints and contributing metadata to the catalog is in the public domain and the catalog itself is available for free reuse and redistribution under a dmoz-like license. Disclaimer: I work for Bitzi.
At the recent ISMIR conference in Indiana, I saw a demonstration of a very similar thing from the Franhoffer (sp) people - it was pretty cool. Just to clarify some of the questions posed by above comments:
1) No tampering is done to the audio - ie there is no watermark, it is "just" signal processing.
2) The system I saw could take any 3 or 4 seconds, so that means that a fingerprint was calculated over the whole song. This means they must have some clever algorithms to make sure that the hash is calculated using the same time slices (or something...)
3) The song has to be in the database. So that means that the fingerprint has already been calculated, and probably had metadata assigned to it either by hand or using mp3 id3 tags (this is a guess). The fingerprint size is about 16KB per song, which seems pretty reasonable.
4) The technique only works on a per-recording basis, so even the same performer doing a slightly different version or another recording won't match if that recording was not already fingerprinted.
5) The version I saw was standard PC software, using audio input through the microphone.
My suspicion is that this technology is more likely to be of value to copyright holders looking to automatically identify violations (eg public airings, radio stations not paying royalties) than it will be useful to joe sixpack (or even people like us...)