Automatic phonetic transcription of large speech corpora: a comparative study