Artificial intelligence for language recognition

Machine learning is developing rapidly at the moment, and this is a relatively sophisticated application that’s already up and running.
Dragonfly Edward Abraham

Edward Abraham

Founder & CEO

Kōkako uses machine learning to monitor the proportion of Māori language broadcast on New Zealand’s iwi radio stations.

Dragonfly is a company based in Wellington, New Zealand which is built around statistics and data science. Abraham describes Dragonfly as having, "an analytical skill set and an appetite for software development".

Te Māngai Pāho is an agency that promotes Māori language and culture through funding initiatives in music, radio, television and new media. Only 2% of New Zealanders converse fluently in Te Reo Māori and Te Māngai Pāho are trying to increase this statistic.

The agency funds radio stations that broadcast more than 10.5 hours of Māori language content per day. Having people monitoring 21 iwi radio stations by ear is not an efficient use of their time or funds. Te Māngai Pāho requested Dragonfly to build an application that could automate this labour intensive monitoring process.

Dragonfly went ahead and built Kōkako on the Catalyst Cloud. Named after a rare New Zealand songbird, the cloud native application uses artificial intelligence to listen to all radio stations simultaneously and determine whether the language being spoken is Māori or English. It then allows radio stations and Te Māngai Pāho to visualise the information through a very intuitive interface.

How did the Catalyst Cloud help Dragonfly in this business venture?

Dragonfly chose the Catalyst Cloud to hold their data. It established hourly transfers of iwi radio from Ngā Taonga Sound and Vision (the flim and media archive based in Wellington) into the object store. Having the Catalyst Cloud in use meant that Dragonfly has saved money with data transfer, by not having to transport large audio files to cloud services overseas.

The audio files are large, and it’s a real advantage that they can be processed in New Zealand rather than going to an international provider – faster and significantly cheaper.

Edward Abraham Founder & CEO at Dragonfly Data Science

Abraham says Catalyst’s infrastructure also makes it easy to host the intensive computation tasks that Kōkako relies on. Kōkako uses a database server, the cloud object-store, and containerised micro-services to process the data. "We trained the machine learning algorithm to distinguish spoken Māori from English, and the trained algorithm is run over new audio as it arrives".

“We use the Catalyst Cloud to keep our servers running online 24-7, so our clients always have access to up to date reporting. We have full control over the servers, but Catalyst looks after all the infrastructure.”

The cloud makes running Kōkako relatively simple and removes a whole layer of responsibility from us.

Edward Abraham Founder & CEO at Dragonfly Data Science

Edward also sees plenty of opportunities for the technology to develop in new areas, such as classifying streaming video. “We’re especially interested in the development of video monitoring in fisheries and whether there are opportunities to help with processing the hours and hours of footage that gets recorded.”

Read more about Kōkako, developed by Dragonfly for Te Māngai Pāho.

In June 2016 Edward Abraham, the Founder and CEO from Dragonfly Data Science, was invited to talk at a Catalyst Cloud breakfast in Auckland about how Dragonfly’s Kōkako technology makes use of our cloud computing platform. Watch Edward's talk on Kōkako at the Cloud Breakfast: