Speaking up for Free TTS

The FreeTTS text-to-speech engine was built by the Speech Integration Group of Sun Microsystems Laboratories. It is written entirely in Java, the popular programming language that was invented by Sun.

It is based on another open-source project, Flite, which was developed at Carnegie Mellon University to run on small devices like PDAs (personal digital assistants). Flite itself is derived from other open-source projects, Festival Speech Synthesis System and FestVox, developed at UK’s University of Edinburgh and Carnegie Mellon University.
 
In the late 1700s, Hungarian scientist Wolfgang von Kempelen created the first speaking machine that was capable of saying whole words and short sentences. His was a mechanical device that consisted of bellows, resonant chambers, reeds and levers, to simulate the human vocal tract. Kempelen’s machine is preserved to this day in Munich.
 
In the 1970s, computers took over the role of modelling the vocal tract to produce synthesized speech, but the results still sounded mechanical. Creating natural sounding synthesized speech requires the modelling of prosody, speaking rate, pitch and so on. The latest approach uses concatenate synthesizers, and do not generate the speech from scratch. Instead, they are derived from speaker recordings. In these synthesizers, many hours of speech are recorded in studios, then chopped up into smaller individual distinctive sounds called phonemes, and indexed.
 
For example, IBM’s latest Naxpress Synthesiser contains about 10,000 samples of each of the 40 or so phonemes in the English language. This sample overkill is because each phoneme is pronounced differently depending on the context on which it appears. This is really the key to natural sounding speech.
 
Continue to read this article on The Malaysia Star ( Malaysia)