INTERSPEECH 2006 - ICSLP
Ninth International Conference on Spoken Language Processing

Pittsburgh, PA, USA
September 17-21, 2006

Six Approaches to Limited Domain Concatenative Speech Synthesis

Robert J. Utama (1), Ann K. Syrdal (2), Alistair Conkie (2)

(1) CAIP, USA; (2) AT&T Labs Research, USA

This paper (this work constitute Robert Utama’s master thesis in the Electrical and Computer Engineering program in Rutgers University) describes the creation of 6 limited-domain Text-to-Speech (TTS) systems that are constrained to digit string and natural number domains (cardinal numbers only). Unit selection-based concatenative TTS systems were implemented in MATLAB to fulfill this goal. We evaluate and discuss various factors that can influence the naturalness or overall quality of the synthesized voice. Some of the factors studied are the length and type of the synthesis unit and the extent of co-articulation represented in the recorded speech database. In the end, we show that it is possible to create a high quality limited domain TTS system either with maximal or with carefully controlled minimal effects of co-articulation.

Full Paper

Bibliographic reference.  Utama, Robert J. / Syrdal, Ann K. / Conkie, Alistair (2006): "Six approaches to limited domain concatenative speech synthesis", In INTERSPEECH-2006, paper 1047-Wed3BuP.9.