A Comparison of Voice Controlled and Mouse Controlled Web Browsing

Results

The experimenters analyzed the task completion times, error rates, and subjective satisfaction of each participant using Microsoft Excel.

Completion Time

A single factor analysis of variance (ANOVA) was performed on the task completion times for each treatment. The results show that navigating slide shows with a mouse is faster than navigating with voice commands. This result was significant at alpha = .05 (f = 4.93, p = .011). Task completion times for navigating hierarchical menus with a mouse were also faster than when navigating with voice commands. These results were statistically significant at alpha = .05 (f = 12.82, p = .00003). However, when navigating tiled maps, the results were not statistically significant at alpha = .05 (f = .27, p = .76), and the null hypothesis cannot be rejected in this case.

For the tasks with statistically significant results (slide show and hierarchical menus), a set of three paired t-tests were performed. The results show significant differences between mouse and both types of voice browsing, but no significant differences between the two voice treatments.

Tables 1a - 1d, summarize the statistical analysis results of the completion times.

Figure 1 shows the means of completion time along with error bars representing the standard deviation in each case.

Error Rates

Mouse errors were excluded from the error rate analysis. A paired t-test was performed on the two voice treatments. With respect to the number of missed commands, the results were not statistically significant at alpha = .05. Hence, the null hypothesis cannot be rejected in this case.

Tables 2a - 2c, summarizes the statistical analysis results of the number of missed commands for each treatment.

Figure 2 shows the means of missed commands in the two treatments: voice-text and voice-numbers.

With respect to the number of misinterpreted commands, the results obtained showed that the number of misinterpreted commands is negligible. Because of the low error rates, no further statistical analysis was warranted.

Table 3 summarizes the means of the number of misinterpreted commands for each treatment. Figure 3 contains the same information in graphical format.

Subjective Satisfaction

A single factor ANOVA of the subjective satisfaction questionnaire shows a preference for text links over numbered links. Results for all three questions were statistically significant at .05 (f = 15.97, p = .000004, f = 20.34, p = .0000003, and f = 15.78, p = .000005, respectively). A set of three paired t-tests were performed. In terms of overall reaction to browsing style and navigating to the desired page, the results show that the voice-text treatment has a statistically significant higher rating than voice-numbers treatment. However, in terms of how easy to use the tool, there was no statistical significance between the two treatments.

Tables 4a - 4d, summarizes the statistical analysis results of the responses to the questionnaire. Question 1 asked users for their overall reaction to the browsing style. Question 2 asked users how easy they found navigation with a particular tool. Question 3 asked users how easy they found it to use the tool. All questions were graded on a scale of 1 to 9, where in question 1, a 1 signified "frustrating" and 9 signified "satisfying", and in questions 2 and 3, a 1 indicated "difficult" and 9 indicated "easy".

Department of Computer Science: Direct questions and comments to the student editorial team