Implementing a Rule-Based Speech Synthesizer on a Mobile Platform

摘要

This paper describes the structure of a Finnish speech synthesis system developed at the University of Turku and evaluates the preliminary results of its implementation and performance on a platform with limited computing power. A rule-based approach was selected due to its high adaptability, low memory and computational capacity requirements. The speech synthesis system is written in Java™ MIDP 2.0 and CLDC 1.1. The synthesis is implemented on Nokia 6680 mobile device as a 65 kilobyte MIDlet. The system produces artificial speech at the sampling rate of 16 kHz. The results show that for a second of synthesized speech it takes 2.66 seconds for the system to produce it. Although the implementation was successful, improvements are needed to achieve a more acceptable level of time consumption.