Abstract. This paper describes the CHIL 2007 evaluation data set provided for the Rich Transcription 2007 Meeting Recognition Evaluation (RT07) in terms of recording setup, scenario, speaker demagogic and transcription process. The corpus consists of 25 interactive seminars recorded at five different recording sites in Europe and the United States in multi-sensory smart rooms. We compare speakers’ talk-time ratios in the interactive seminars with lecture data and multi-party meeting data. We show that the length of individual speaker’s contributions helps to position interactive seminars between lectures and meetings in terms of speaker interactivity. We also study the differences between the manual transcription of narrow-field and far-field audio recording.