Computer speech synthesis is a complicated process. Phoneticians, people who study sound in language, study exactly what influences which sounds we hear. Pairs of sounds such as "p" and "b" are very similar, and the distinctions between vowels is often subtle. Before a computer could synthesize speech, the scientists had to study what makes a human hear a "b" instead of a "p," or the sound "eh" instead of "ah".
Analyzing Text
The actual process of computer speech synthesis begins by turning the text the computer is given into sound units called phonemes. A phoneme is the smallest distinct unit of sound, such as the "c" sound at the beginning of cat. The computer has rules telling it how combinations of letters interact. For example, it will know that the "e" at the end of "came" makes the "e" into a tense vowel, instead of sound in "cam". The program will also have rules for phrases and sentences. For example, it will know to pause briefly after a comma, and to raise the tone at the end of the sentence.
Creating Speech
The computer next takes the complex phonological information which is information about speech sounds and turns it into waveforms. There are different ways of doing this. Some speech synthesis programs actually have a stored library of different speech sounds, which they can combine into words. Others have a sophisticated model of the human vocal tract. The second approach is often less "real" sounding, but it offers more flexibility. It allows for changes in parameters to make the voice male or female, older or younger, or higher and lower in pitch.
