Transcript pptx
Introduction and Motivation Speech recognition is cumbersome Document navigation and editing tasks rarely studied Commercial tools support two navigation methods “Move down five lines, move right four words” Go to The quick brown fox SLOW! FRUSTRATING! ERROR-PRONE! We say “Let the computer do the walking…” Expert Interview Experience with speech recognition Satisfaction with speech recognition Suggestions for improvements Comments/opinions on our proposals Navigation Proposals 1. Move down five lines, move right four words 2. Go to The quick brown 4. Auto-scroll Go down… faster… faster… slower… stop Go right… faster… stop fox 5. Auto-scroll and pause 3. Go to page 3, paragraph 2, sentence 5, word 8, character 4. at natural landmarks: section headers, paragraphs, and userdefined landmarks: middle of page, topic sentence Navigation Proposals (Moving within a page) 1 2 3 4 5 6 7 8 9 6. Hierarchical Keypad Grid 7. Sunray 1 2 8. X-and-Y 7 3 4 5 6 Pilot Study Navigate document Forward Short, medium, long Pre-interview Post-interview Video taped 3 Modes Keyboard+Mouse, SR, SpeedNav™ Medium Then, highlight short stretch of text 3 Documents 4 subjects Non-impaired Novice SR Motor-impaired Novice SR Non-impaired SR expert Motor-impaired SR expert Backwards Familiar Read once Unknown How do we ensure that subjects’ familiar documents are matched? Will document retention affect repeated task performance? Evaluation Metrics Navigation Errors Undershoot Overshoot Inappropriate scroll speed Highlighting Errors Start position incorrect #chars, words, sentences Time to completion Number of commands spoken Number of words spoken Subjective approval Training time Fatigue End position incorrect #chars, words, sentences Command recognition errors Should application record timestamps of user to make timing easier? Measure long distance (between screen) navigation technique separately from short distance (on same screen) navigation? Design Questions Cursor Default scroll in Word leaves cursor at bottom -- can’t read what’s below the screen! Scroll with cursor on left margin, right margin, or center? How might position affect ability to scan left and right? Cursor movement draws the eye; keep cursor stationary and move document instead? Sunray Option: Can people follow diagonal cursor movement? Speed Need various starting speeds and speed multipliers. Usercontrollable? Speech recognition adds one second delay! Causes overshoot! User must anticipate when to stop.