AI system
- System
-
Te Hiku Media te reo Māori automatic speech recognition and transcription tool (Papa Reo platform; trained with Mozilla DeepSpeech-era open-source tooling, later NVIDIA NeMo)
- Model
-
Not assessed
- Release
-
Not assessed
- Application developer
-
Te Hiku Media (Kaitaia, New Zealand) - Role
-
Not assessed
- Configuration
-
Not assessed
- Inputs and outputs
-
Not assessed
- Autonomy
-
Not assessed
- Workflow
-
A community-contributed read-speech corpus (310 hours in ten days) held under Te Hiku Media's data-sovereignty rules. Bootstrapping from open-source speech models and, later, NVIDIA's NeMo toolkit.
- Human oversight
-
Not assessed
- Relation to AI
-
The accuracy figures come from Te Hiku Media. MIT Technology Review reported 86% in 2022. NVIDIA, a tooling vendor, and Te Hiku's chief executive through RNZ stated 92% in 2024. The University of Auckland's independent test compares outputs against human transcripts, with results to follow. No cited source reports an outcome for any speaker, learner or archive beyond the developer's own broadcasting use.
- AI contribution
-
The accuracy figures come from Te Hiku Media. MIT Technology Review reported 86% in 2022. NVIDIA, a tooling vendor, and Te Hiku's chief executive through RNZ stated 92% in 2024. The University of Auckland's independent test compares outputs against human transcripts, with results to follow. No cited source reports an outcome for any speaker, learner or archive beyond the developer's own broadcasting use.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
transcribes te reo Māori with 92% accuracy
- Category
-
Improved Task Accuracy
- Status
-
Not assessed
- Beneficiary groups
-
Organizations. Society.
- Beneficiaries
-
Not assessed
- Reach
-
Not assessed
- Distribution and access
-
Not assessed
- Duration
-
Not assessed
- Expected?
-
Yes
- Costs and harms
-
Not assessed
Measurement
- Studies
-
Not assessed
- Study design
-
Not assessed
- Measurement
-
Not assessed
- Setting and sample
-
Not assessed
- Measures
-
transcribes te reo Māori with 92% accuracy
- Comparison
-
Assumed alternative: te reo Māori speech recognition attempts by major tech companies
- Results
-
Assumed comparison. Result: an accuracy of 92%.
- Additional value from AI
-
Not assessed
- Limitations
-
Not assessed
- Denominators
-
Not assessed
- Materials for replication
-
Not assessed
Money
- Kind of value
-
Not assessed
- Amount
-
Not assessed
- Currency and prices
-
Not assessed
- Period and unit
-
Not assessed
- Realization
-
Not assessed
- Recipients and payers
-
Not assessed
- Gross or net
-
Not assessed
- Included costs
-
Not assessed
- Comparison
-
Not assessed
- Method
-
Not assessed
- Return and payback
-
Not assessed
People and places
- Participants
-
Developer:
Te Hiku Media (Kaitaia, New Zealand). Operator:
Te Hiku Media; test partners Archive of Māori and Pacific Sound (University of Auckland), RNZ and Ngā Taonga Sound & Vision. - Jurisdictions
-
Benefit: Kaitaia and Auckland,
New Zealand. Developer:
New Zealand. Operator:
New Zealand. - Ownership and control
-
Not assessed
- Dependencies
-
Not assessed
- Cross-border connections
-
Not assessed
- Geopolitical implications
-
Not assessed
Follow-up
- Conditions for replication
-
Recordings resemble the training data. The 2024 tests deliberately probe regional dialects, recording quality and sung forms such as mōteatea and waiata. Archive holders license recordings on the partners' conditions (no pecuniary gain, previously broadcast material only).
- Replication attempts
-
1 follow-up entries. Outcome unresolved: 1.
- Later findings
-
11 September 2024. RNZ announced on 11 September 2024 that it is integrating Te Hiku's Papa Reo API into its workflows for search and captions, with RNZ and Ngā Taonga supplying archival te reo broadcasts to train and review the tools.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0319-source-0001–0005
- Citations
-
Numbered references below.
- Language
-
Not assessed
- Source relationships
-
Operator, Independent, Developer
- Original evidence
-
Not assessed
- Access and versions
-
Not assessed
- Supported findings
-
Not assessed
- Evidence links
-
Not assessed
- Evidence strength
-
Not assessed
- Missing evidence
-
Not assessed
Grades
- Assessed benefits
-
Organizations. Society.
- Depth
-
Not assessed
- Reasons
-
Organizations: Te Hiku Media and its archive partners report the tool performs well for broadcasting and are testing it on archival material. No measured time saving or completed archive transcription is documented. Society: Benefits to te reo speakers and learners are the stated aim; no cited source measures them.
- Observed or intended
-
Not assessed
- Rubric
-
Not assessed
- Assessor and date
-
Not assessed
- Alternative assessments
-
Not assessed
- Fill
-
Independently reported
Entry details
- Kind
-
Not assessed
- Related entries
-
Not assessed
- Field
-
Language technology / Indigenous language revitalisation. Iwi media and audiovisual archives.
- Purpose
-
Built to transcribe te reo Māori recordings so that archives and broadcasts become searchable and readable by learners, under Māori data-sovereignty rules. The transcription tests are that intended use.
- Dates
-
Date basis unassessed: 2024 (precision unassessed). Updated: 22 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: bf2371c20b4c…
References
- Other identifiers
-
Not assessed
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
8 revisions. Latest: 22 October 2026.
Sources
-
News report Independent author
Hao K (2022). A new vision of artificial intelligence for the people. MIT Technology Review, 22 April 2022.
Te Hiku created its very first te reo speech recognition model with 86% accuracy.
-
Press release Deployer
University of Auckland (2024). Putting te reo artificial intelligence to the test. News, 18 March 2024.
This ever-evolving tool is performing very well for their purposes, providing their audience with an alternative way of exploring their output.
This project will see a sampling of recordings put through the automatic transcription process, then the output compared against human-generated ones
-
Press release Developer
Lee A (2024). Māori Speech AI Model Helps Preserve and Promote New Zealand Indigenous Language. NVIDIA Blog, 16 January 2024 (NVIDIA supplies the NeMo toolkit and hardware).
Te Hiku Media's automatic speech recognition model transcribes te reo Māori with 92% accuracy using trustworthy AI and the NVIDIA NeMo toolkit.
-
Press release Deployer
RNZ (2024). Media collaboration to develop Māori language tools. RNZ media release, 11 September 2024 (RNZ is a data partner of Te Hiku Media in this agreement).
a review of the tools to check for improved handling of dialect nuances and the accuracy of transcription, in particular for names no longer in use.