The text to speech goes through speech dispatcher which by default probably uses espeak. I believe there are better text to speech engines like festival which you should be able to have speech dispatcher use quite easily but I've never tried this.