Commanding and Re-Dictation: Developing Eyes-Free Voice-Based Interaction for Editing Dictated Text

What is it about?

We investigated the natural user behavior of making eyes-free speech-based corrections in dictated text. Two major behaviors emerged — by using commands, e.g., "DELETE <phrase>" (including conversational-styled commands, e.g., "I would like to delete <phrase>, please"), and by re-speaking over erroneous portions of the text, e.g., correcting "he go there", by saying "he goes there". We developed two eyes-free voice-based techniques to support each of the two identified behaviors and compared them within a controlled setting. Results showed that while Re-dictation was faster and more efficient for making longer and more complex corrections in the text, Commanding (especially command-based deletions), was the more preferred technique for making one or two-word edits in the text. Further, while re-dictation gave the users a sense of perceived naturalness of use, commanding provided a sense of perceived control. Based on these observations, we propose an interaction technique that modelessly combines the two individual techniques for eyes-free correction of text. We evaluated our proposed technique in a realistic setting, results from which showed improvements in the performance and usability of the combined technique over either of the two techniques used alone.

Featured Image

Why is it important?

Towards an envisioned interaction paradigm that better integrates computing into the users' everyday actions and mobility, voice interaction is likely to play a major role as voice input provides a fast, natural and device-independent modality of interacting with computers. Further, voice input can be performed hands- and eyes-free, leaving the users' whole body free to engage in other tasks. A core everyday computing task is text input/editing. Prior research has established that text entry with voice is 3 to 5 times as fast as typing on a physical/virtual keyboard. Yet, the effective text entry rate is limited as speech recognition is far from perfect and almost 80% of the users' input time is consumed in correcting errors in the dictated text. Our current research tries to address this research gap and takes a step towards establishing a feasible, real-time voice-based error correction technique for editing typed/dictated text.