ON INTEGRATING TONAL INFORMATION INTO CHINESE SPEECH RECOGNITION