Joint acoustic factor learning for robust deep neural network based automatic speech recognition.