Human Language Technologies (HLT) – Text-to-speech Systems

In order to provide timely information in an affordable way for projects such as OpenPhone and Lwazi, text-to-speech (TTS) systems are required that automatically transform digital text into speech. Such systems have only been developed for a limited number of languages internationally, and prior to 2004 no freely available text-to-speech system existed for any indigenous South African language.

In 2004, in partnership with the Local Language Speech Technology Initiative (LLSTI) we developed and evaluated a general domain isiZulu TTS system based on the open source Festival Speech Synthesis System. The isiZulu voice (“Buhle”) was released into the open domain. An Afrikaans voice soon followed, as well as a Setswana voice, developed as part of the OpenPhone project.

Our current focus involves the building of similar systems for each of the 11 official languages of South Africa as part of project Lwazi. Voices currently released are built using concatenative TTS technology. Research focuses on the efficient development of systems when limited resources are available, including optimal text selection strategies and rapid bootstrapping of audio segmentation labels. In future, we hope to see the results of our intonation modelling research become available for integration in our TTS systems.

Links

TTS-cubed: a collection of speech synthesis tools and modules that extends the Festival, Flite and FreeTTS speech synthesis systems.

Buhle isiZulu voice

Setswana OpenPhone TTS corpus: segmented and labelled voice recordings

Selected publications:

D.R. Van Niekerk and E. Barnard, “Important factors in HMM-based phonetic segmentation”, In Proceedings of the 18th Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp 25-28, Pietermaritzburg, South Africa, November 2007.

J.A. Louw, M. Davel and E. Barnard, “A general-purpose isiZulu Speech Synthesiser”, South African journal of African languages, 2006.

J. A. Louw and M. Davel, “Halfphones: A Backoff Mechanism for Diphone Unit Selection Synthesis”. In Proceedings of the 17th Annual Symposium of the Pattern Recognition Association of South Africa, pp. 42-45 Parys, November 2006.

E. Barnard and M. Davel, “Automatic error detection in alignments for speech synthesis”, In Proceedings of the 17th Annual Symposium of the Pattern Recognition Association of South Africa, pp. 53- 56 Parys, November 2006.