Exploiting deep neural networks for detection-based speech recognition