If it already knew, using 16000 can listen to the correct speed to the voice, but the commands that contain more than one word are cut and points cannot be listened, it can be due to the fact that when it calls to this command or to this minute and exact second in the one that is the command, a parameter passes to know that length has in time to reproduce it 2 seconds or 10 seconds according to the length of the sound to reproducing.
I think that to change the voices is something more complex and till now we do not have tools for it: (