Using n-best recognition output for extractive summarization and keyword extraction in meeting speech