Search for ko¯kako goes hi-tech
Artificial intelligence software that can detect the call of the South Island ko¯ kako is being hailed as a revolutionary new tool in the search for the elusive bird, of which there have been only two verified sightings in the last 50 years.
Chris Blackbourn has developed software that isolates bird vocalisations in audio files. It is able to listen for ko¯ kako calls among hundreds of hours of recordings.
A computer game programmer, Blackbourn responded to an appeal from the South Island Ko¯ kako Charitable Trust to help with analysing birdsong to identify the presence of specific birds.
The last verified sighting of the South Island ko¯ kako in the 20th century was in 1967, and the species was officially declared extinct by the Department of Conservation in 2008. However, after a 2007 sighting near Reefton was accepted as accurate, DOC upgraded the bird’s status to ‘‘data deficient’’ in 2013.
Since January 2017, the South Island ko¯kako, known as the ‘‘grey ghost’’, has been the most wanted bird in the country, with the trust putting up a $10,000 reward to anyone who can provide photographic evidence of it in the wild.
Blackbourn created a neural network that classifies New Zealand birds based on their sound patterns,
Inger Perkins, South Island Ko¯kako Charitable Trust using the vocalisations of 50 species, including both the North Island and South Island ko¯ kako.
South Island Ko¯kako Charitable Trust general manager Inger Perkins said Blackbourn’s software was revolutionary. It meant the trust could record more audio to be analysed, and process older files, which had previously been a daunting task.
‘‘The fact we now have the ability to do that is a giant leap forward for bird-searching mankind.’’
More than 400 hours of recordings thought to contain ko¯kako calls, recorded in the Landsborough Valley in South Westland in 2011, were analysed in just a few days. Nearly 118,000 bird vocalisations were identified, almost 21,000 of which were suitable for further processing.
The hundreds of hours of recordings were cut down to 10 minutes of audio suitable for further review. It found that 89 of the vocalisations sounded more like a ko¯ kako than any other bird species.
When the calls were checked by ear, the differences were more apparent and were thought to be more like tu¯ ı¯ than ko¯ kako.
Blackbourn said the system would continue to be refined, but he was confident that if South Island ko¯ kako calls were present on a recording, the software would identify them.
Perkins said the trust was hugely grateful to Blackbourn for his work.
‘‘Although we have not identified the South Island ko¯ kako in this recording data, we are thrilled to now have the capability to use computer systems and AI to process large quantities of audio data.
‘‘We are optimistic that by sharing the search widely, asking the public to be our eyes and ears, and by using the latest technology, we’re giving this bird the best chance to be found and conserved appropriately.’’
For the first time, the South Island ko¯ kako is in the 2019 Bird of the Year competition. Voting closes on November 10.
‘‘[It] is a giant leap forward.’’