Speaker-adaptive learning of resonance targets in a hidden trajectory model of speech coarticulation