Hybrid HMM/ANN Systems for Speech Recognition: Overview and New Research Directions