NeXTeNS stands for `Nederlandse Extensie voor Tekst naar Spraak', or `Dutch Extension for Text to Speech'. The projects aims at developing a modern, clean, multi-platform, open source text-to-speech system for Dutch that is freely available for research and education purposes.

Background

Apart from being an legitimate research topic in itself, text-to-speech (and speech synthesis from annotated input) is a valuable tool for developing and testing linguistic and phonetic theory and technology. Areas of interest include prosody, spoken dialogue systems, concept-to-speech, and multi-modal output employing e.g. talking heads.

Unfortunately, the current availability of TTS for Dutch is rather limited, and dominated by commercial TTS systems limited to the MS Windows operating system. In general, commercial systems are tailored towards the demands of end users who wish to use the system as a black box for voice output, rather than for the purpose of research and development. Moreover, the fact that the source code of commercial systems is unavailable forms a major drawback for their use in research and education. At the same time, the TTS systems developed at Dutch universities are hampered by limited support and documentation, suffer from incomplete, messy, or ad hoc platform-dependent code, and and a lack of support for state-of-the-art applications.

Yet, developing a new TTS system for Dutch from scratch would require a major investment in terms of time and finance. It also appears to be a waste of effort, at least in part, as many of the knowledge and resources underlying the outdated systems for Dutch can in fact be reused. In addition, generic, multi-lingual, and freely available TTS architectures have recently been put forward. In particular, the Festival system and the Mbrola diphone synthesizer appear to have become the de facto standard for open source TTS systems in the research and education community.

Goals

The goal of the NeXTeNS project is to develop and deliver a modern, clean, multi-platform, open source TTS system for Dutch. For its general architecture and basic facilities, it relies on the Festival system. For waveform synthesis, it makes use of the Mbrola diphone synthesizer and its nl3 database. Furthermore, it incorporates knowledge and resources from the KUNTTS system (developed at the University of Nijmegen), as well as state-of-the-art NLP tools for Dutch grapheme-to-phoneme conversion, POS tagging, accent placement and prosodic phrasing (developed at the University of Tilburg).

The final system will minimally have to meet the following criteria:

• converts unrestricted ASCII text into high quality Dutch speech encoded as an audio file
• runs minimally on a PC platform with either MS Windows or Linux
• usable in interactive mode by means of the OS shell's command line or a Scheme command line interpreter, as well as by various API's (Scheme, C/C++) and in TCP/IP server mode
• open source code, and freely available for research and education
• reasonably well documented for both end users and developers
• supported by means of a central web site

Implementation

The initial participants in the NeXTeNS project are the University of Nijmegen (Department of Language and Speech) and the University of Tilburg (Induction of Linguistic Knowledge group).

The project relies on substantial financial support by Stichting Spraaktechnologie (Foundation for Speech Technology), a Dutch organization for the promotion of speech technology.

The NeXTeNS project has started in February 2002 and will run for at least one year.

Contact

We hope that the NeXTeNS project will attract participation and contributions by others. If you are interested in joining the NeXTeNS project, please contact Erwin Marsi.