[transmit101]

This would be a real money spinner, if it could be made to work. The problem, assuming you are using an analysis algorithm to attempt to extract the harmonic information, is that you are attempting to solve the problem of polyphonic transcription, which is one of the holy grails of computer music research - and is definitely an unsolved problem, especially when the analysed song is complex or unpredictable (jazz chords, hendrix tunings etc.), or when there is a lot of noise, as is the case with almost all popular music.

It is quite straight-forward to use a Fourier transform to discover the energy contained at certain audio frequencies, but extracting musical structure from this data is terribly difficult, partly because of the perceptual ambiguities in the human hearing system, and partly because it is basically impossible to confidently separate the identity of the constituent instruments within the audio spectrum. What instrument emitted this or that harmonic at 1800hz? The answer is crucial to identifying the chords, and nobody really knows how to reliably and programatically determine that, on an arbitrary recording, just yet...

[kolis]

You are right that this task is not solved yet. You may read about our technology at http://www.music-ir.org/mirex/2008/abs/cbms_cover_song_id.pd...

Our tests showed that 92% of the time we show correct chords, which is obviously not a 100%.

If you will have a time, give it a try, it works really well.