radioman - we certainly can clean up the audio files eventually (background noise reduction in the female files would be another big one, along with shortening the silences), but for 2.0 we're just trying to get rid of most of the glaring inaccuracies rather than getting everything to sound perfect. The syllable-by-syllable feature as I've said before was really just meant to keep people from thinking that the audio feature is broken / doesn't work because they're not hearing any audio for a particular word (just as we've now added UniHan built-in so that people stop thinking the handwriting recognizer is defective because when they accidentally tap on a really rare character the dictionary doesn't have an entry for it) - if we wanted to do a proper text-to-speech system that would take a lot more work (and would probably require several recordings of each syllable/tone combination in order to capture different lengths, pitches halfway between mid/high, etc).