Since our results are based on only 67 subjects and 8 tokens, we cannot consider them to be decisive - more data would be needed for that. But, they do show us trends that are, in our belief, very important. They show that speech synthesis is not a solved problem. As individuals get older and rely more on the telephone, that lifeline is effectively not serving them as it should. They understand speech that is less natural, but the synthetic speech shows a pronounced decrease in understanding.
We also note that some information is easier to understand than others; it would seem that our ability to predict items in a semantically restrained domain remains constant as we age.
We have anecdotal evidence about hearing problems. In general, people who admitted they had hearing problems did better then those who did not (but whose spouses had dragged them over to our experiment saying they really needed to participate, implying their hearing was not very good). This implies that those who admit they have a problem and do something about it will continue to understand better than the others.
We should also note that the people who attend Carnegie Mellon Homecoming are a specific group. They have obtained higher education and are very mobile. It is evident that the largest part of the elderly population who desperately need to use the telephone for essential services are not as highly educated and, especially, are not as mobile (would not have physically have been able to attend Homecoming). We therefore need to extend and validate our data by repeating the experiment in activity centers and assisted living communities for the elderly.
Finally, we can see that the clues that our speaker used to try to be better understood were not completely successful. But we can measure the elements of prosody (timing, intonation and phrasing) that she employed and conduct further studies that use other combinations of these elements in order to find which elements make synthetic speech more understandable.