Machine learning methods are ideal for classifying streaming data rapidly. We developed Kōkako, a language monitoring app to identify different languages broadcast on radio stations in New Zealand. Kōkako won a Best Design Award in 2016.
Tracking the proportion of Māori and English spoken
Te Māngai Pāho is a government agency charged with funding and monitoring the te reo Māori (Māori language) content of iwi radio. Radio stations are funded to deliver more than 10 hours of Te Reo content in a day which, for compliance purposes, is often delivered in specific blocks of time. Broadcasters, however, report that this inhibits a natural use of the language and drives away audiences that might enjoy a more bilingual style.
An automated system
Te Māngai Pāho was looking for an automated and more flexible system to replace their labour intensive random manual monitoring system. We used machine learning methods (in this case recurrent neural networks) to develop a system that classifies radio as either being te reo, English, or music. The classification is continuous (made many times a second), and the system follows speakers as they flip between languages coding Māori in red, English in blue and music in purple.
Fast and accurate language identification
The recurrent neural network achieves more than 95% accuracy in detecting each language and music, and can process three hours of audio in less than a minute. A set of manually labelled data was required to train the system, but once trained, the system is able to keep classifying new data as it arrives. This machine learning technique is well suited to other audio, video or streaming data classification tasks.
Partnership with Salted Herring
We were pleased to work with Wellington digital design agency Salted Herring on this project, who designed the user interface and created the brand identity. We worked together on the front-end development. Kōkako was named a Supreme Winner in the interactive category of the 2016 Best Design Awards and awarded a Purple Pin.
Watch a Native Affairs, Maōri TV interview about this project.
The team at Dragonfly have been great to work with. They quickly understood what we were trying to achieve and came up with a very innovative solution.
In fact, their prototype attracted a lot of interest nationally and internationally before we even had a chance to put it into operation.
Dragonfly took an idea and turned it into something pretty special. They made this project happen for us.
Thomas Hood
Manager Corporate Services
Te Māngai Pāho .