Neural Networks to Model Guitar Amps

  Currently there are many guitar "modellers" available to recreate sounds via DSP of various well known amps - but they never sound or respond "just right".
  How about using Neural Networks to come up with accurate transfer functions to process straight guitar input signals to realistic sounding and responding amplifier output signals? You train the networks by playing the guitar through the actual amp you want to model, and use backpropogation of error to define the specific amp transfer fucntion over some training time.