“Some people are starting to say we don’t want to be relying on overseas technology for everything,” he said.
“When you’re talking about Māori language and Māori knowledge, that’s too important to hand over.”
The copyright of Māori voices and data sovereignty remained central concerns, he said, particularly as overseas companies faced scrutiny over how personal information was handled.
“We have to be careful with our knowledge and our data so that we don’t get to that stage of redress,” he said.
A dialect voice takes shape
It is within that broader vision that a new artificial intelligence voice – designed to speak in the Waikato-Maniapoto dialect of te reo Māori – has emerged at the University of Waikato.
Developed at the University of Waikato by Masters student Kingsley Eng under the supervision of Keegan, the AI-generated voice aims to preserve regional dialects and strengthen rangatahi (youth) engagement with te reo.
The project gained momentum after Keegan received an unexpected email from Google offering an unsolicited funding award to support responsible Māori language technologies – a message Keegan initially thought might be a hoax.
The unrestricted backing allowed the team to move quickly from concept to development, engaging specialist support and Māori language experts.
Keegan said the work came as technology shifted toward voice-based computing, making high-quality spoken language tools increasingly important.
The system’s pronunciation has been tuned to what he described as the “Māori ear”, with the Waikato-Maniapoto model designed as a template other iwi could adapt to build their own dialect-based voices in future.
Still in testing but nearing public release, the project is being hailed as a major step toward “sovereign AI” – technology built by Māori, for Māori – as voice-based computing becomes increasingly central to everyday life.
Training a voice for the ‘Māori ear’
The AI voice was trained using recordings from a Māori language expert from the Waikato-Maniapoto region, whose work teaching and promoting te reo Māori is recognised locally and nationally.
Developers recorded about 10 hours of her speaking a day – roughly 2200 sentences – which were then broken down by the software into small sound units to learn how she pronounces different parts of words.
Keegan said the result was so realistic that even people closest to the speaker struggled to tell the difference.
Dialect over standardisation
Keegan said dialect-specific tools were essential to preserving the richness of te reo Māori, rather than relying on a single standardised form of the language.
“When we’re talking about technology, having tools that can speak different dialects is really important,” he said.
“That’s where the richness and beauty of te reo comes through – things you don’t always get if you’re only using a general, common Māori language.”
The copyright of Māori voices and data sovereignty was incredibly important, as was ensuring their voice model was trained correctly, he said.
Keegan was aware of overseas incidents involving tech companies allegedly accessing personal data.
“We have to be careful with our knowledge and our data so that we don’t get to that stage of redress.”
Trust versus big tech
The team of academics at the University of Waikato were up against some of the most powerful companies in the world, all with deep pockets, Keegan said.
“The big companies have not millions, not even hundreds of millions, but they have billions of dollars, and they have hundreds, maybe thousands of engineers working on them.”
Keegan said he believed the creation of AI in New Zealand by teams of academics was more “trustworthy”.
The project has been years in the making, driven by Keegan’s long-standing focus on the future of te reo Māori in digital spaces.
He said the idea of building a dialect-based voice had been sitting with him for a long time, but dedicated funding allowed the team to move from concept to reality by bringing in specialist support, including Eng and Māori language experts.
Keegan described the process as both technically demanding and deeply personal, combining advanced speech modelling with cultural responsibility.
Rather than simply creating another synthetic voice, the goal was to produce a system that reflected the nuances and identity of Waikato-Maniapoto speakers, ensuring rangatahi could hear technology speak in ways that felt familiar.
The work also highlights the growing role of the University of Waikato’s AI Institute, where multiple projects are exploring how artificial intelligence can support language revitalisation and community-led innovation.
Keegan said he often found himself thinking late into the night about the next steps for Māori language technology, driven by a belief that the digital world must evolve alongside the needs of indigenous communities.
While awards and recognition have followed, he said the real impact lay in creating tools that made te reo Māori more visible – and audible – in everyday life.
Beyond one project
Keegan said the dialect voice project was just one part of a much wider programme of artificial intelligence research under way at the University of Waikato, with multiple initiatives being led through the AI Institute.
“Through the AI Institute, which Albert Bifet leads, there’s a whole range of other really interesting projects happening as well.”
Director of the AI Institute at the University of Waikato Alfred Bifet said the institute, launched in 2021, built on decades of AI expertise at Waikato, including the Weka software developed in the 1990s.
Weka is an open-source machine-learning tool that enables researchers to analyse large datasets and train models to recognise patterns and make predictions, and is freely available under a public licence.
“It has 20,000 citations, more than five million downloads,” Bifet said.
Tom Eley is a multimedia journalist at the Waikato Herald. Before he joined the Hamilton-based team, he worked for the Weekend Sun and Sunlive. He previously worked as a journalist in Canada for Black Press Media and won a fellowship with the Vancouver Sun.
