Synthesis of speech

Abstract

A text-to-speech (TTS) synthesis should contribute to the pleasantness,

intelligibility, and speed of speech-based human-machine interactions.

The initial step is that articulatory speech synthesis attempts to mathematically

model the human vocal system through area functions of its vocal organs. In this

project, the author has used a male vocal system and designed a large unit

inventory practical. Next, the design of the units is acoustically driven but limited

by a complex pronunciation model.

Finally, concatenative speech synthesis links different length of prerecorded

speech samples together that were obtained from natural speech. Fortunately,

concatenative synthesis requires less computational complexity at the expense

of larger memory space.

A set of vocabularies and a prerecorded speech corpus containing 40 phonemes

serves as the basic acoustical units were created by the author before testing

and carrying out the experiment for the TIS system. Further research was done

on how other methods like Pitch-Synchronous Overlap Add (PSOLA) can better

concatenate the chain of sound elements and eventually produce a continuous

speech in English.