Acoustic Model Adaptation with Multiple Supervisions