The Input

• The input stream was first broken down into 1/20th of a second samples which were fed to a finite Fourier transform function to extract frequency spectrum information.
• Blocks of 30 seconds were assembled and considered individualy.
• Each block was again fed to a fourier transform function, again with respect to time to determine 'beat' frequencies across the spectrum.

• Final input vector was a 24x32 matrix (24 sample points for pitch and 32 for rhythm/tempo), or a vector of 768 values.