In the first experiment, all the utterances are regarded as `overlapping.' The subjects are eight men. First we prepared a lecture-style speech, and subjects were asked to listen to it and say any questions or comments. Because the subjects are not familiar with the topic, the subjects rarely barged in on it.
We also asked them to do casual conversations. Now they were familiar to the topic, so they made responses easily. In case the overlap is small, the speech merged by the server seemed very natural. When several long utterances are overlapped, however, they found difficulty to understand what the speakers are saying.