Extract some of the most common low-level audio feauters.
Estimate Poisson-distributed Hidden Markov Models.
Train some Self-organizing maps.