Version: (using Devel) Installed from: Compiled sources kde 4.2 rc If Phonon output is selected, KttsMgr pronounces just few fragments of the text, if the output is set to Alsa, KttsMgr works fine.
I can reproduce this behavior with KDE 4.2. Alsa also works fine. But when I switch back from alsa to Phonon, Phonon output currently works. This lasts until kttsd is restarted. KttsMgr also shows the same problems when choosing "Speak clipboard contents". The sentences are always spoken from the beginning to some point. It seems to me that this point is some percentage of the sentence length. Of long sentences a longer part is being spoken. Short (one word) "sentences" get mostly completely skipped. If I select "Keep audio files" the resulting files are always in the correct length. Maybe this bug is related to the issues in Bug 157810 which has been fixed for KDE 4.2
My guess is that this is related to a bug in xine libraries which affects short wav files, see e.g. http://bugs.debian.org/418059 Try to verify whether you're using the xine backend. If yes, try to play any short wav file using dragon player. If you encounter the same problem, then the bug is probably not within kttsd. (If you have flite installed, you can easily create a short wav file by running: flite -t "Short text" -o example.wav)
I think your guess is right. I'm using the xine backend I just tried to play some wav files with dragon player and get the same behavior there. But if I start dragonplayer within bash with $1 as filename ("dragon example.wav") the whole wav is played. If I choose to replay it afterwards, it gets skipped. I also tried Amarok (1.4) because it also uses the Xine-Engine and the wav files were played correctly with Amarok.
Closing as an upstream bug, if you have reason to believe this is an error within KDE components feel free to reopen it. Bug has probably been fixed with libxine version 1.1.90 and above (under development).
Reassigning to phonon - for the case it's not a problem in libxine.