O n f e r e n c e

Short paper 
Background 
Text-to-Speech (TTS) synthesis is a relatively new and relatively little used addition to 
CALL materials. A review of literature on the use of TTS synthesis for educational 
purposes showed the 1990s as the decade which promised most hope for its future. 
Stratil, Burkhardt, Jarratt and Yandle (1987) were positive for the future of TTS synthesis 
in teaching as far back as the pre-nineties.

---

-105- 
2014 CALL Conference 
LINGUAPOLIS
www.antwerpcall.be 
They noted a positive reaction from students and teachers at the time but, despite the 
very significant strides made in the development of TTS systems in the intervening 
quarter century, this anticipated potential has not been widely exploited to date. TTS 
has not had a great deal of interest among second/foreign language teachers and has not 
been widely used in CALL (Gupta & Schulze, 2012). Various reasons for this have been 
proffered - less than perfect prosody; lack of clarity and consistency; voices being 
insufficiently expressive; lacking in human emotion; only small range of voices available 
(Delogu, Conte, & Sementina, 1998; Keller & Zellner-Keller, 2000; Sha, 2010). Higgins 
(1988) concluded that one should not bother using synthesised speech as a model for 
learners because it cannot account for the complexity of human language. While there 
have been huge advances in the TTS technology since the 1980s it is still less than 
perfect and some claim that the basic criticism still holds valid (Handley, 2005). 
If Computer-Assisted Language Learning (CALL) programs are to be developed where the 
learner can have a genuinely interactive relationship with the computer so that learning 
can become highly personalised and tailored to the unique needs of each individual, then 
TTS synthesis comes centre stage. Speech recognition technology also has an obvious 
role but it is difficult to envisage at present how it may deal with recognising the 
imperfect utterances of learners and this question is not being considered in the present 
paper. 
With the aid of increasingly robust artificial intelligence (AI) systems computers can 
interact with the learner in relatively unpredictable ways. It would therefore not be 
possible to pre-record novel utterances which the computer may produce. The question 
arises as to whether present day TTS systems are ‘good enough’ to be part of a 
personalised CALL program which offers the learner greater control over his/her own 
learning process. This paper describes an evaluation of an Irish language TTS system, 
ABAIR
, which is being developed at the Centre for Language and Communications 
Studies, Trinity College, Dublin, in the context of its being used in a specifically designed 
interactive game-like CALL environment. 
The general framework on which the 
ABAIR
system is built is Festival 1.95, which is a 
version of the Festival Speech Synthesis System (Clark, Richmond, & King, 2004) 
developed at the Centre for Speech Technology Research (CSTR) at the University of 
Edinburgh. Festival offers a general framework for building speech synthesis systems. It 
is an open source system, from which a number of speech engines are available for the 
development of new voices using unit-selection synthesis. Unit-selection synthesis is 
based on recordings of real human speech. It is a form of concatenative speech synthesis 
where potentially minute segments of recorded speech are pieced together in order to 
produce synthetic speech output. This is one of the most popular techniques in use to 
date, particularly by commercial companies, as it generates the most natural sounding 
speech output.