Papa Reo — Te Hiku Media's Māori Speech-AI Platform Built on Indigenous Data Sovereignty

Te Hiku Media, a charitable media organisation collectively owned by five Far North iwi (Ngāti Kuri, Te Aupōuri, Ngāi Takoto, Te Rarawa and Ngāti Kahu), ran the Kōrero Māori crowdsourcing campaign in 2017, collecting over 300 hours of labelled te reo Māori speech in just 10 days from more than 2,500 participants reading over 200,000 phrases.

In 2019 New Zealand's Ministry of Business, Innovation and Employment invested NZ$13 million to launch Papa Reo, a seven-year, multilingual natural-language-processing platform — notable as the only project of its kind in New Zealand not led by a university.

Using the open-source NVIDIA NeMo toolkit and NVIDIA A100 GPUs, Te Hiku Media built automatic speech recognition that reaches 92% accuracy for te reo Māori and 82% for bilingual Māori-English speech. The models power the Kaituhi transcription service and the Whare Kōrero archive (roughly 1,000 hours of content spanning three decades), which is used by around 20 Māori radio stations.

Central to the project is the Kaitiakitanga License, which keeps Māori speech data under Māori governance rather than the public domain or corporate control — securing elder support and transparent consent before any data collection began. The model is now cited elsewhere as a template for indigenous-led data sovereignty in language AI.

Caution: Papa Reo's continuity depends on a single seven-year government grant and a small specialist team; no independent, post-grant sustainability evaluation was found in public reporting.

Read the full analysis: https://tehiku.nz/te-hiku-tech/papa-reo/14135/te-reo-maori-speech-recognition