This writer for the Washington Post has a problem with the personae of in-car navigation devices, and with speech recognition systems in general. The article a little over the top in its opinions because the writer is trying to be humorous, but it fairly points out that the speech output of devices can become annoying very quickly if not done properly.
Note that the problem as identified isn't necessarily with the sound of the voice, it's the lack of context machines have when giving directions, and the fact that the problem reveals itself over time, as in a long car trip. These things are often not tested well, or tested at all. The article also identifies lack of "naturalness" as a problem with speech systems, but trying to achieve an undefined kind of "naturalness" causes problems as well.