Reviving Te Reo Māori: Unleashing the Power of AI for Language Preservation and Promotion

What is Te Reo Māori?

There is a danger to indigenous languages. According to UNESCO, one out of every two weeks, or almost 3,000, could vanish before the century ends. This represents three-quarters of the total. Te Hiku Media, a broadcaster based in New Zealand that specializes in the indigenous language of the Māori people called te reo, is utilizing reliable AI to aid in the preservation and revitalization of the language as part of an effort to safeguard such languages.

Te Hiku Media is creating ASR models for the Polynesian language te reo by employing open and honest ways of gathering and analyzing speech data in order to uphold the Māori people’s data sovereignty. The NVIDIA A100 Tensor Core GPUs and the open-source NVIDIA NeMo toolbox for ASR were used to construct the speech-to-text models, which accurately transcribe te reo with a rate of 92%. With an accuracy rate of 82%, it can also transcribe bilingual speech in English and Te Reo. The Māori people have created these essential tools to help them tell and preserve their tales.

Read: 10 AI In Energy Management Trends To Look Out For In 2024

Te Hiku Media- content distribution platform

With the goal of giving te reo a voice on the radio, Te Hiku Media launched over 30 years ago. Incorporating television broadcasting into its operations over the years, the group met with community elders in 2013 to formulate a plan for content sharing in the digital age, in response to the proliferation of the internet. Te Hiku Media opted to create its own content distribution platform rather than submit its video and audio sources to well-known worldwide platforms, which require users to sign away certain rights associated with the content in their t******************* of usage.

The website, which translates to “house of speech” in Te Reo, currently contains digitized archival material spanning over 30 years. It features over 1,000 hours of te reo native speakers, including some who were born in the late 19th century, as well as more current content from second-language learners and bilingual Māori. Approximately twenty Māori radio stations currently utilize Whare Kōrero and submit their programs there. Members of the community get access to the material via a mobile app.

The team at Te Hiku Media rapidly recognized the immense importance of this collection for language revitalization efforts; yet, manual transcribing demanded a significant amount of time and energy from their limited resources. To speed up its work utilizing ASR, the firm started its trustworthy AI activities in 2016.

Read: 10 AI In Manufacturing Trends To Look Out For In 2024

In order to gather speech data in an open, ethical, and community-focused manner, Te Hiku Media first sought out the cooperation of elders by describing its purpose, collecting their phrases aloud, and inviting them to the station. Kōrero Māori was a crowdsourcing initiative that Te Hiku Media launched to gather speech samples with high levels of labeling in accordance with the Kaitiakitanga license. This license guarantees that the data will only be used for the benefit of the Māori people. The te reo Māori ASR models were built and trained using over 300 hours of labeled speech data, which was provided by over 2,500 individuals who signed up in just 10 days to read over 200,000 phrases.

The ASR module for speech AI from the NVIDIA NeMo toolkit is currently used across Te Hiku Media’s pipeline, along with other open-source trustworthy AI products. Pretrained models for language model construction are part of the NeMo toolkit’s neural modules, which are building blocks. The ASR models developed by Te Hiku Media power Kaituhi, an online te reo Māori transcribing service that is presently accessible. Native Hawaiians and Mohawks in southeastern Canada are currently engaged in ASR projects that are inspired by these efforts.

Read: Amazon’s Revolutionary Retail Revealed: Meet Rufus, An Astonishing AI Conversational Shopper