Introduction

A text to speech synthesiser is a system whereby user inputs text and obtains synthesised speech in return.  

Speech synthesisers are used in various capacities such as computer voice, in mobile devices, as voice-mail and in-car navigation system.  Generally there are 3 approaches to computer speech synthesis :

1. Articulatory synthesis - uses the computational models of the articulators and their movements to synthesise speech.
2. Rule-based synthesis/formant synthesis - a source-filter speech production method.  The vocal tract filter is constructed from a number of resonances similar to the formants of natural speech.  One of the most widely known rule-based systems is the MITalk from MIT.
3. Concatenative synthesis - is the process of synthesis by concatenating sub-speech units to form speech.  It has become popular of late due to the advances in speech modification techniques, and the increasingly affordable cost of computing capabilities and memory.

This research attempts to answer some of the questions developers have to answer in the course of building a concatenative Text To Speech (TTS) synthesis systems.

1. How to obtain a good set of speech units for synthesis.  This includes synthesis of unseen context speech units.
2. How to synthesise speech that is perceived to be natural, as per the input text.
3. How to ensure smooth concatenation of speech units.

Publications

P.H. Low, S. Vaseghi, “Synthesis Of Unseen Context And Spectral And Pitch Contour Smoothing In Concatenated Text To Speech Synthesis”, ICASSP, Florida, USA, Vol. 1, pp. 469-472, 2002

P.H. Low, S. Vaseghi, “Application of Microprosody Models In Text To Speech Synthesis”, ICSLP, Denver, USA, pp. 2413-2416, 2002

Author/Research student  : Phuay Hui LOW

Supervisor : Prof. Saeed VASEGHI

Last updated : 25 Feb 2003