Preliminary results using non-discriminative transformations confirm the effectiveness of this approach, and suggest discriminatively trained models should be sufficient for indexing of end-user videos recorded with hand-help devices, transcription, summarization, or translation of content in tele-presence systems, or other applications.