Speech-based interfaces

VoiceNotes: A Speech Interface for a Hand-Held Voice Notetaker, by Lisa J. Stifelman, et. al., in Proceedings of INTERCHI, Amsterdam, The Netherlands, Apr. 24-29, 1993, pp. 179-186.

Oops, sorry, no points on this one. It's been a tough week...

Good points

Bad points

Designing SpeechActs: Issues in Speech User Interfaces, by Nicole Yankelovic, et. al., in Proceedings of CHI '95: Human Factors in Computing Systems, Denver, CO, May 1995, pp. 369-376.

Good points Great paper!

  1. Recognition rates were a poor indicator of satisfaction. Something to remember for UIs in general. I would like to read something about what does increase user satisfaction in UIs in general.
  2. Disconnection from GUI techniques. I really like their approach to interaction: "Today, you have..." provides an indicator of the current context without it being tedious or even noticable.
  3. The elimination of "dialog boxes." Yay! One wonders how applicable their observation is to UI's in general...

Bad points

  1. The justification for the research was somewhat lame. Traveling professionals who feel out of touch?
  2. Addition of "directive prompts." If the user doesn't provide the desired response, then presumably it's because the user thinks the system already knows what should be done. A better approach might be to make a best guess and inform the user, either then or later.
Empirical Evaluation of Interactive Multimodal Error Correction, by Bernhard Suhm, in Proceedings of the IEEE Workshop on Speech Recognition and Understanding, Santa Barbara, USA, Dec. 1997.

Good points

  1. It's great that speech combined with textual error correction can in theory get 78 wpm. (But see below).

Bad points

  1. It's not at all clear exactly what they were measuring, and how the different correction techniques were applied. For example, how was "choice from N-best" applied? How did it differ for the different techniques (spell vs write...).
  2. The predications in Table 4 seem rather suspect. 78 wpm is pretty fast just to type -- and this includes error correction? I would have liked to see more justification for the predictions.

John Reekie, February 27th, 1998.