NZ-based charitable media and technology organisation, Te Reo Irirangi o Te Hiku o Te Ika (Te Hiku Media), Radio New Zealand (RNZ) and the Ngā Taonga Sound & Vision audiovisual archive have entered into a new collaboration to further develop bilingual transcription and speech recognition for te reo Māori and New Zealand English.
In a first for the public media organisation, RNZ has agreed to supply archival radio broadcasts to the iwi-led charitable trust to support its groundbreaking natural language processing tools. Ngā Taonga, the audiovisual archive of Aotearoa, which cares for the RNZ archive, is providing support to ensure the recordings are cared for and handled appropriately.
Natural language processing is a branch of artificial intelligence (AI) that uses machine learning (ML) technology. Machine learning is the science of developing algorithms and statistical models to perform complex tasks such as automatic speech recognition (ASR).
Te Hiku Media, as part of its Papa Reo data science project, has been developing tools that can transcribe te reo Māori, give pronunciation feedback, and turn text into speech. This has expanded to include NZ English and work has begun on sister languages to te reo Māori, such as, ‘ōlelo Hawai’i. The goal of using the RNZ material is to train the tools to recognise a wider spectrum of speech, dialects and accents.
Te Hiku Media CEO Peter-Lucas Jones said: “Our tools already outperform attempts by major tech companies at te reo Māori ASR, with an accuracy of 92%. This accuracy and overall quality are directly related to the quality of the data, the quality of our relationships, and the Māori language and data science capabilities of our team. This collaboration reflects all of these things, with our three organisations spending time building trust and understanding in our relationship and demonstrating our commitment to quality te reo Māori outcomes and data sovereignty.”
Te Hiku Media has been recognised globally as an advocate for Indigenous data sovereignty and community-led data science. Ngā Taonga and RNZ have put in place some strict parameters in terms of Te Hiku Media’s use of the archival material.
RNZ regularly shares its content with other media partners, but this collaboration is the first time it is granting access to its content to help develop AI tools.
RNZ Chief Technology Officer Mark Bullen said it was very important for RNZ to partner with organisations that shared its value of te reo Māori as a taonga when considering the use of AI.
“RNZ is particularly interested in tools that help unlock our online audio, supporting improved search, discovery and accessibility in the form of captioning of content. This will support both our internal production capabilities and RNZ’s audiences. And indeed, part of the collaboration with Te Hiku Media allows for RNZ to have access to these tools. But we also felt that the values of Te Hiku Media, their well-documented position on data sovereignty and their support of whānau in preserving their oral histories aligned with our values and our responsibilities under RNZ’s Charter.”
Ngā Taonga Tumu Whakarae (Chief Executive) Honiana Love said the collaboration with RNZ and Te Hiku Media was an extension of some of the early testing the archive did with te reo Māori transcription tool.
“We definitely saw merit in the tool given our own cataloguing work and the number of requests we have for te reo Māori transcriptions from whānau. We care for the largest body of the historical recordings of te reo Māori and mātauranga Māori in the world and in recent years have made a commitment to cataloguing te reo Māori material in te reo Māori. Supporting this work to help recognise a wider spectrum of speech, dialects and accents is important to our work to protect taonga Māori.”
RNZ and Te Hiku Media signed a Memorandum of Understanding in December 2023, agreeing in principle to give access to RNZ’s recordings. Ngā Taonga, which holds the archival kōrero, then led a consultation process with iwi/Māori stakeholders.
Following this consultation process, which included a webinar as well as direct engagement, the three parties have agreed to take the following points into account as part of the collaboration:
- there should be no pecuniary gain from the access to the recordings;
- the material supplied is only that which has been previously broadcast and has no restrictions;
- the taonga cannot be rebroadcast or reused in any form other than what has been agreed to;
- 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.
The first tranche of archival material being delivered is a selection from the already publicly available He Rerenga Kōrero collection. RNZ is in the process of integrating the Papa Reo API into its workflows, initially adding transcribed audio to the search capability and then allowing it to add captions to its player in the near future.
Visit https://www.rnz.co.nz/