Consideration of Class Labels for Japanese End-to-End Speech Recognition