For years, those of us working in bioacoustics have been searching for patterns in the sounds around us. What kinds of sounds do different animals make? What are the differences, and what is their function? To answer these questions, we record vocalisations, analyse them, and play back the same or modified sounds to see how animals respond.
There is always a limit to how much we can do. You can only record so many sounds – which can be particularly challenging with wild animals – and although much of the analysis is automated, there is always a degree of manual checking involved to verify that everything has been coded correctly.
Until now, that is. Last week I came across a new study in which 60 million (!) bird sounds had been analysed. That is a completely different order of magnitude from anything we have managed to analyse so far. The researchers obtained their data via BirdNET, an online database of bird sound recordings. Anyone can upload their recordings there, which naturally generates a wealth of data. They then used a machine-learning algorithm to analyse all 60 million recordings. In this particular study, they found that birds in areas with more artificial light started singing 18 minutes earlier and continued for 32 minutes longer than their counterparts in less illuminated areas. We already knew, of course, that light (and noise!) affects the rhythms of animals, but it had never been investigated on such a large scale before.
I look forward to seeing what the future holds in this field
Even when we want to focus on just a single species, AI is having an enormous impact on the way we can analyse vocalisations. In a recent study on chickens, in which only four distinct vocalisations had previously been recognised, researchers have now identified 42 different types of sound. The automated method of analysis revealed more clusters than had previously been thought to exist. Whether chickens experience things in the same way is, of course, another question entirely – but it does show that there is still a great deal left to discover. I look forward to seeing what the future holds in this field. We are already using bioacoustics to automatically monitor how many species occur in a given area, to assess the state of biodiversity, and to track the natural rhythms of animals. But there are also many possibilities over the longer term. We may yet find universal patterns across species that we had never previously considered. Perhaps we will soon be able to automatically identify individuals by their sounds, making it simpler to map social networks. Or perhaps it will help with welfare monitoring, allowing us to measure more quickly and efficiently whether an animal is experiencing problems.
Either way, machine learning is clearly transforming the way we can analyse sounds. I am keen to see where the limits lie. As far as I am concerned, the future of bioacoustic research simply sounds fantastic.
Michelle Spierings is a behavioural biologist at the Institute of Biology Leiden and the University of Vienna, where she works on the development of speech, song and cognition in human infants, common marmosets, budgerigars and zebra finches.