AI saves the day for musicians (Score 1)

As a researcher in pattern recognition who also manages a band, this topic holds particular interest for me.

I haven't read the article but it would surprise me if this method was more effective than a frequency/power spectrum (the distribution of sound energy over all frequencies) or even a basic neural net classifier (both mentioned by previous commentors) at stylistic classification of music. Fractal dimension reduces the whole waveform to a single number; a frequency/power spectrum contains a lot more information. (For the fractally challenged: a straight line has a dimension of 1, and a solid plane a dimension of 2; a music waveform has some fractional dimension between 1 and 2... it's roughly a measure of the regular "squiggliness" of the waveform. Fractal dimension is not necessarily a measure of self similarity as is implied by the news brief; it's just that the self similar patterns that we popularly call "fractals" have an interesting fractal dimension, like 2 2/3 or something.)

I think music recognition technology is the key to resolving the conflict between artists who want to make money off their recordings and fans who want to sample a wide variety. It's a tough business folks; Metallica doesn't deserve much sympathy but most bands are extremely exploited by record companies and deserve to make as much money as possible off their art. Someday releasing a record will also entail releasing a host of net-bots that look for unauthorized, publicly available copies of the music.

Stylistic classification is not that hard. Much harder is reliably recognizing a particular piece of music (i.e. creating a bot that scours the net, looking for copies of "Master of Puppets"). It's easy to fool a frequency power spectrum classifier, for example, by tacking on some tones to the end of the track, or reequalizing the track, or adding some low frequency inaudible noise.

This is where AI steps in. The biggest feather in the cap of AI is the technology used in 95% of all speech recognition systems: Hidden Markov Models (they are probabilistic versions of deterministic finite state automata, for those geeks out there who have been subjected to the torture of a class in computer science theory). The same technology can be used to identify pieces of music.