Is a speech recognizer useful for characteristic analysis of classroom lecture speech?