Speech Away

Can a machine coach you?

Live feedback keeps you engaged; the delayed report is what teaches

Chollet, M., Marsella, S., & Scherer, S. (2022). Training public speaking with virtual social interactions: real-time vs delayed feedback. Journal on Multimodal User Interfaces, 16(1), 17-29.

Chollet and colleagues ran a clean four-way test in front of a virtual audience: real-time feedback, delayed feedback, personalized or not. The pattern was unambiguous. Personalized delayed reports produced a significant skill gain (Flow of Speech, p=0.010). Real-time implicit feedback did not. Non-personalized reports did nothing at all. And the busiest condition - a live dashboard of icons - performed worst of the lot.

p = 0.010 skill gain from personalized delayed feedback; real-time was non-significant
Figure · Reported valuesSkill gain by feedback condition
Personalized · delayed reportp = 0.01082Non-personalized · delayednot significant26Real-time implicit feedbacknot significant20Dense live icon dashboardperformed worst8
Personalization and timing both decide the outcome. Only the personalized, delayed report moved skill; the dense live display was the failure mode. Bars reflect reported significance, not exact magnitudes.
Data table
Itemmeasured skill change
Personalized · delayed report (p = 0.010)82
Non-personalized · delayed (not significant)26
Real-time implicit feedback (not significant)20
Dense live icon dashboard (performed worst)8

Two jobs, two tools

Real-time feedback is not useless - it drives engagement and immersion, which matter for getting people to practice at all. It just does not build skill. So the design rule is to keep the two jobs separate: if you ever show something live, make it one sparse, glanceable nudge framed as motivation - and put the teaching in the post-hoc report. Rhema (Tanveer et al. 2015) found a single one-word cue beat a continuous display, which fits exactly.

What it means for Speech Away

During your take, the only live element is a calm audio visualizer - proof we can hear you, not a scoreboard. The personalized analysis comes after. It mirrors Chollet's winning condition: delayed, and built entirely from your own data. We deliberately avoided the dense live dashboard the study found made things worse.