Statistical Parametric Speech Synthesis is a type of [vocal-synthesis] that uses statistical modeling and/or machine learning to act as a [tts] engine. This is an alternative approach to [concatenative-synthesis] which uses recordings of phonemes or small snippets of sound to produce speech.
The best overview of this that I've found online is here: https://wiki.aalto.fi/display/ITSP/Statistical+parametric+speech+synthesis archive link: https://web.archive.org/web/20210312033816/https://wiki.aalto.fi/display/ITSP/Statistical+parametric+speech+synthesis It gives a truly excellent overview of the state of the art, detailed descriptions of various projects and how they work, and the benefits/drawbacks of different approaches.