One year on from the start of the pandemic, we remember how the hum and roar of our morning lives were replaced by a symphony of bird sounds. These chirpy sounds of aggressive posturing and mate attraction used to go unnoticed by many before lockdown simplified our urban soundscapes.
It's spring again and the music is back. Birdwatch Ireland launched a new podcast to coincide with the uptick in spring bird activity. In Your Nature is full of interesting stories that help you make sense of the sounds of the flying dinosaurs that share your neighbourhood. As bird activity hots up for breeding season, the podcast helps listeners untangle the cacophony of voices. There is a sense of achievement in pulling out the song of the blackbird or great tit from the background chirpings and cheepings.
We can teach computers to beat humans in chess, but it has taken a surprisingly long time to get effective birdsong recognition. There is the problem of recognising a large number of individual species, with about 10,000 named species of birds worldwide. This is relatively straightforward to solve; with a big enough library of bird sounds, artificial intelligence algorithms can be trained to match sounds to names. The range of possible matches can be reduced by limiting the solutions to bird species that are likely to occur in the location of the sound recording.
A more complex problem is taking whole soundscapes, such as the dawn chorus, with many species singing at the same time, and pulling out individual songs for recognition, often against a background of car sounds, planes flying overhead, humans talking and laughing and dogs barking. This is “noisy” soundscape data. Humans, with some training, can learn to identify individual species from the background noise.
Tuning in
Professional ecologists and ornithologists can be extremely good at tuning in to particular species, even in noisy environments, with many bird surveys done by sound alone. A new computer algorithm, BirdNet, has been developed by scientists working in the US and Germany, which takes soundscape recordings and matches the recorded birdsong to nearly 1,000 of the most common bird species in North America and Europe. Acoustic monitoring and computer classification can potentially add to the huge amount of human recorded bird data, which is used to better understand bird populations and conservation needs.
The Deep Artificial Neural Network computer models developed in BirdNet use data on bird songs, from online libraries of bird sounds, to "learn" how to associate song data with bird identities. Multiple recordings of each species were used to cover the whole repertoire of a species and capture regional bird "accents". These models were also trained with common sources of false detections, things such as humans whistling, footsteps, speech, urban background noise, wind and rain. The treasure trove of sounds in Google audio set was used for these non-bird sounds. An aside – you can listen to more than 120 examples of toothbrush sounds here; the mind boggles.
Sneaky starling
After training, the neural net model was tested against independent data to see how well it performed, including complex soundscape data that had been annotated by professional ornithologists to extract individual species from overlapping songs. The results were very encouraging, although some birds, such as the European starling, which mimics other species and even human sounds, remained a significant challenge. Background noise also made it more difficult to identify bird species from recordings. You can try it out yourself by downloading the BirdNet app and using it to identify your local birds.
Birds can be hard to see if they are hanging out in the undergrowth or high in trees, but being able to recognise their calls gives us an insight into the soap operas of their lives. It can be tricky to learn birdsong, but apps such as BirdNet and podcasts such as In Your Nature make it much easier and more enjoyable. We are not yet at the point of a Dr Dolittle translator that lets us talk to the animals, but the first step is surely listening to what the birds tell us.
Yvonne Buckley is an ecologist, Irish Research Council laureate and professor of zoology at Trinity College Dublin