Can a machine coach you?
A virtual coach worked - but only the personalized post-hoc report
Hoque, M. E., Courgeon, M., Martin, J.-C., Mutlu, B., & Picard, R. W. (2013). MACH: My Automated Conversation coacH. UbiComp 2013, 697-706.
MACH put a virtual human on a laptop to run mock interviews, then tested whether it actually helped. Ninety participants, three conditions, and judges who never knew who was in which group. The result drew a sharp line: the people who got a personalized post-hoc report - their own video plus behavioral analytics - improved. The people who just practiced, or just watched generic advice, did not.
Data table
| Item | measured improvement |
|---|---|
| Post-hoc video + analytics (personalized feedback) | 84 |
| Watched advice video (no significant gain) | 22 |
| Practice alone (no significant gain) | 18 |
Why post-hoc and personalized#
Two ingredients did the work, and both matter. Post-hoc: the analysis came after the speech, where the learner could absorb it - not during, where it would just distract. Personalized: it was their data, their patterns - not generic tips anyone could have written. Strip either ingredient and the effect vanished.
What it means for Speech Away#
Speech Away is built on exactly this finding. The heavy lifting - the scorecard, the structure mirror, the one fix - lands in a report after your sixty seconds, generated from your actual transcript, timing, and audio. We resist the temptation to throw a busy live dashboard at you mid-speech, because the best-controlled trial in this space says that part does not teach.