Digital Signal Processing Reference
In-Depth Information
Figure 4-5.
Correct rate of the active speaker determination. The experimental setup was: four
speakers S1, S2, S3, and S4 sat in different corner of the car; microphones were placed in the
car in four configurations; and the speeds of car were maintained for three conditions (idle,
city, and highway).
4.2
Statistics of the interactions in an MSDS
In order to determine the interactive behavior in a multiple speaker
dialogue, the thirty-two testers were divided further into eight groups for the
second experiment. Each group of testers performed multi-speaker dialogues
six times. Each tester was given the freedom of deciding whether he wanted
to join in the dialogues. That is, during a multi-speaker dialogue, the number
of parties could be one, two, three, or four. Also, the testers were free to
choose one of the three task domains, i.e., navigation guide, weather
information, or stock price information. Within this paradigm, we computed
three statistics about the interactions of an MSDS. The first was how many
parties engaged in each MSDS. The second was the percentage of interaction
types that occurred in the MSDS. The last was the rate of correct interaction
type determination.
