John McKenna is interested in the application of statistical, data-driven, and machine-learning techniques to the processing of speech. More specifically, he is interested in speaker characterisation and voice transformation.

Dr. McKenna's current work is best described as Automatic Speaker Characterisation, that is, the filtering of speaker-characteristic information from the speech signal using digital signal processing and machine learning techniques. This has enormous potential for the robustification and flexibility of both automatic speech and speaker recognition, and speech generation systems. Automatically characterising speakers would allow removal of speaker variability from speaker-independent ASR (Automatic Speech Recognition) while attaining the quality of speaker-dependent systems. Multilingual, multi-speaker, and multistyle speech synthesis have been identified as important trends in speech generation systems. With recent advances in data-driven learning, automatic speaker characterisation techniques are necessary in order to collect data for these applications.

Dr. McKenna's approach to Automatic Speaker Characterisation is to initially obtain automatic separation of glottal source and vocal-tract filter using analysis architectures that employ stochastic filters.

Dr. McKenna is also a member of the Irish Speech Group (ISG) and its recent emergence hints at exciting times ahead for Speech Technology in Ireland. The ISG currently comprises speech researchers and linguists in seven Irish institutions. Its aim is to develop speech technology for the Irish language, while acting as an inter-institutional forum for cooperation and collaboration. All resulting applications and resources, e.g. text-to-speech (TTS) and ASR of Irish, will be freely available to both users (e.g. educators, disabled) and developers. The Irish Speech Group hopes that speech technology in Irish will enjoy the same availability as other popular languages. It has established a partnership with Welsh language developers and has been successful in a research bid to the tune of 700K to collect data and build a first-generation Irish TTS system.