Could music fingerprinting be copyright infringement?

Music fingerprinting is the process of extracting essential information from digitized audio recordings (e.g. CD tracks) so it can be searched in a database. Typically a large database is compiled of known recordings and then the database is searched for fingerprints of unknown audio samples. When a match is found, the previously unknown audio sample can be identified reliably. Music fingerprinting has a lot of applications, some of them are:

• Song identification services like Shazam. The user can record a small sample of a song (played on the radio or as background music in a shop) using a mobile phone, which is then fingerprinted. The server matches this fingerprint against a database of fingerprints of known songs. Then the server provides the artist and song title, so the user can purchase a CD of that song.
• Automatic royalty collection from radio stations. Songs played by a radio station are fingerprinted and then identified automatically.
• Filtering of infringing audio from file sharing services of video sites like Youtube.

Music fingerprinting is big business, but entry into the market with independently developed software (not just buying one of the existing solutions) is not simple. And this is a gross understatement. There are several reasons why it is so hard to enter the market:

• Music fingerprinting algorithms are patented and optimizations of these algorithms in existing applications are protected by trade secrets. One must first negotiate a license for all relevant patents and then one has to spend much time to develop the software, also tuning the algorithm for best matching capabilities as these aspects of existing applications are protected by trade secrets. It may be possible to write an algorithm that does not use any of the patented technologies.
• Real life applications of audio fingerprinting require a huge database of fingerprints of (nearly) all recorded music. Shazam has a database of 8 million songs (the equivalent of more than half a million CDs). Before you can compile this database, you need access to the recordings in the first place. Although it is certainly possible to create fingerprints of recordings without making intermediate copies, it is much easier to do it when you can make intermediate copies, especially if the process of fingerprinting must be repeated later. But in most countries these copies cannot be made without permission.
• When you decide to improve the fingerprinting method, you have to repeat the entire fingerprinting process for all recordings.
• The format of fingerprints is not standardized (not even publicly documented). Therefore you cannot buy just the database of fingerprints and use it with your own application.
• The copyright status of music fingerprints themselves is unclear. This risk can be mitigated though.

This article focuses on the copyright implications of music fingerprinting.

How music fingerprinting works

• System and methods for recognizing sound and music signals in high noise and distortion United States Patent: 6990453
• Robust and invariant audio pattern matching United States Patent: 7627477

This article by Bryan Jacobs describes how the matching algorithm in Shazam works. Here is an explanation of the music fingerprinting and matching algorithm that is accompanied by a Matlab implementation. Finally Roy van Rijn in The Netherlands created a rough implementation of a music fingerprinting program in Java as described