A Māori academic has created the world’s first AI-generated voice that speaks Māori uniquely in the Waikato-Maniapoto dialect, and the technology has the potential to include other iwi dialects in the future.
Te Taka Keegan, associate professor at Waikato University’s software engineering department, said as Māori he and his colleagues have long been thinking about how to build their own AI tools and this was the first step in that process.
He said they follow in the footsteps of Northland’s Te Hiku Media which has made trailblazing moves into the AI language tools space.
“This is really kind of supporting what they're doing but providing an avenue where each iwi can provide their own voice.”
Broadly speaking, there are iwi variations within the Māori language. For example, the word ‘whakarongo’ (to listen) in Ngāpuhi dialect is ‘hakarongo’; in Taranaki it’s ‘w’akarongo’; in Tūhoe they say ‘whakarono’; and for Ngāi Tahu it’s ‘whakaroko’.
There are also regional differences predominantly between the east and west - tupuna/tipuna (ancestor), tātou/tātau (all of us), tētehi/tētahi (one of).
Keegan said that rather than having a generic, “pan-Māori” voice that may exist in other AI tools, the technology can create voices that speak in these specific dialects.
“Voices are identity, really, so that’s what this is about,” said Keegan. “Let's have a look at the process to create a voice and let's provide an avenue where if other iwi are interested as well they can produce their own voice.”
He said that in the future there’s a “clear possibility” to revive a voice or a way of speaking too, if there are enough recordings available.
There is also a speech-to-text option that runs through a couple of different systems, including AI, to provide an answer in Māori that the tool can then ‘read’ to speak.
He said from a computer science perspective, keyboards and text messaging “isn’t going to be here forever”.
“Shortly there'll come a time when you'll be speaking to your computer, you'll be speaking to your phone, and it will be speaking back. So, this is in preparation for the day when our devices are speaking back to us, we'll be able to get them speaking back to us not just in te reo Māori, but in the reo Māori dialect that we connect to.
“That's the future – how we interact with our devices.”
Staff and students who tested the tool were hard pressed to distinguish the real voice from the AI-generated.
Māori liaison advisor Hirini Edwards said he found it “very hard” to tell the two apart when he submitted his answers. “I didn't know if I got it correct or not, but after re-listening to all of them just before, I think I got a few wrong.”
Student and Māori language tutor Ariana Stewart said: “The voices sounded so similar so when I first heard it I was like, 'Woah, what is real? What is not?'”
She said with normal AI, she was used to it sounding monotonous and robotic. “So, it's kind of fresh to actually hear a voice that sounds similar to what you would hear at home.”
The morning's headlines in 90 seconds, including an injury blow for the Black Ferns, the main suspect in Madeleine McCann’s disappearance is out of jail, and self driving cars are here. (Source: 1News)
Concerns over ‘colonised’ AI-produced reo and voice 'ownership'
Keegan said existing AI language tools produce “colonised” versions of te reo, providing a “clean version” that struggles to recognise that Māori have “lots of truths”.
“When you talk to an AI tool like Chat GPT, it's only going to give you one answer. It's not going to allow for the multiple truths that we as Māori are comfortable with.
“When is Māori new year, is it Puanga, is it Matariki? We're comfortable with that, but a colonised version of the tool wouldn't be, it would just give you one answer.”
When it comes to te reo and mātauranga Māori, Keegan believes Māori need to take back control of the technology.
"The AI tools we don't have control, we've lost sovereignty over that control, but here's an example how we can gain back, and once we've got this tool working, that's our next step. How can we gain back control of what these AI tools are saying in te reo Māori?"
And when it comes to his own tool, he applies a Māori worldview to intellectual property rights. From a legal, Pākehā perspective, he said, the speaker cedes their right to their voice.
“But from a Māori perspective, ownership isn't the right question, it's more kaitiakitanga – who's the kaitiakitanga of that particular voice? So these are some discussions we are having.”
He believes whānau, hapū and iwi all have a say in the matter because the speaker’s voice is of a community.
“Because this is a Waikato voice, a Maniapoto voice, Raukawa as well, we need to sit down with those three iwi entities and have a talk about how we protect this voice – not so much for her, the person who it sounds like – but for her children and for the generations to come from those particular iwi.
“That's a discussion we haven't finalised yet, but it's an important discussion and it's important that we have that discussion from a Māori perspective rather than a legal, intellectual property perspective.”
SHARE ME