One general approach might be to assign an OscillatorBank to each frequency band. For example, you could take a LiveSpectralAnalysis and feed it into two different SpectrumModifiers: one that selects frequencies above a cutoff, and one that selects frequencies at or below the cutoff. The high SpectrumModifier could feed an OscillatorBank that is panned to one position, and the low SpectrumModifier could feed an OscillatorBank panned to a different position. If the results sound promising you could continue the process, splitting the live signal into ranges of low-to-high frequency, somehwat like an EQ.
Curious to hear how it goes and what you discover!