Speech Synthesis

A major driving force in speech synthesis has come from text-to-speech (TTS). A TTS system assumes that the text already exists in machine-readable form, such as an ASCII file. The machine-readable form is possible obtained from optical character recognition. TTS converts the text symbols to a parameter stream representing sounds. This includes expanding common abbreviations, such as “Pres.,” and symbols like “&”. Also, the system figures out how to handle numbers: 1492 can be read as a date, and even the dollar amount $1492.00 could be read starting with “fourteen hundred...” or “one thousand four hundred....” After creating a uniform symbol stream. The system creates initial sound parameter representation, often at the world level-some words may simply be looked up in a dictionary.
Other parts of the stream are broken down into morphemes, the syntactic basic units of the language. With luck, a group of text symbols corresponds to one morpheme. It is often the case that there is a regular mapping between the symbols of such a group and some sound, in which case the group can be turned into sound. As a last resort, the system converts individual text symbols to sound using rules. The system synthesises sounds from the parameter string based on an articulatory model or using sampled sounds, LPC, or formats. The synthesis system may store units at the level of phonemes, diphthongs, syllables, or words.