This work looks very cool from the demo video. Have yet to read the paper, but the parametric inversion for generating sounds from features seems very intriguing.