Closed vvasco closed 4 years ago
I used a stopwatch to compare the time we compute to a ground truth. I realized that speaking was introducing latency in the computation of the starting time: to trigger the start, the robot pronounces a sentence and we were waiting for the sentence to be pronounced and computing the starting time only after. As a result, the time computed was lower than the ground truth. I improved this in ee33af1, by splitting the sentence in two shorter ones, and avoid waiting when triggering the start. With this change, the temporal metric is reasonable.
We want to be sure that the computed time is not affected by disturbances (such as questions during the test).