Artist uses AI to “deepfake” the dawn chorus

The AI technology behind deepfakes has been used to recreate the urban dawn chorus for an artwork highlighting how birdlife is under threat from 24/7 culture. Contemporary artist Dr Alexandra Daisy Ginsberg worked with AI company Faculty to recreate the sound of the ritual in which thousands of birds from different species sing to greet the dawn in spring and early summer as they defend their territory and call for mates.

This extraordinary natural phenomenon, which some have compared to a “bird opera”, is under threat as the presence of city light and noise pushes birds to start earlier or sing at higher pitch or volume, disrupting their mating patterns and leading to falling populations. In Ginsberg’s immersive sound installation, entitled Machine Auguries, the artist asks how the city might sound with changing, homogenising, or diminishing bird populations.

The piece was commissioned by Somerset House and A/D/O by MINI for the major exhibition 24/7 opening at Somerset House on 31st October.

Ginsberg, who has spent ten years researching synthetic biology and human impacts on nature and who is currently a resident artist of Somerset House Studios, said: “Increasing urbanisation and climate change are altering the dawn chorus as birds are forced to adapt, if they can. Urban birds, such as sparrows, blackbirds and great tits have been found to sing higher, louder, and earlier, using more energy to communicate while their ability to call to mates is diminished. Near airports, researchers have shown blackbirds singing for longer and modifying their song. “Machine Auguries brings attention to the impact on bird populations and explores similarities in the learning of birdsong and the training of artificial neural networks.”

Machine learning expert Dr Przemek Witaszczyk, who worked on the project as part of the Faculty fellowship programme, said: “This project challenges the way we think about our changing world. By generating a synthetic dawn chorus with artificial intelligence, it aims at shocking us into contemplating a future in which we have only digital substitutes for our natural species.”

In the installation, the sound of real birds singing alternates with artificial birdsong generated using a machine learning technique known as a Generative Adversarial Network (GAN). A GAN pits two artificial neural networks against each other, both trained on the same dataset of images, videos or sounds.

The first tries to create new samples that are good enough to trick the second network, which works to determine whether the new samples are real or not. The same technique has been used to generate lifelike fake video and audio – known as “deepfake” – of human beings including well known celebrities and politicians. Faculty is one of the UK’s leading researchers in the field having worked with pro-democracy campaign group the Alliance of Democracies to highlight the threat to democratic elections posed by AI-generated deepfakes of politicians.

Dr Witaszczyk used solo recordings of chiffchaffs, great tits, redstarts, robins, thrushes, and entire dawn choruses to train two specialised neural networks known as WaveGANs. He used a different technique to work out the sequence of different bird songs in a London dawn chorus which he then used to develop a “script” for the AI generated version using a technique known as a stochastic sequence generator.

The generated birdsong clips, which become increasingly lifelike, have been mixed with natural birdsongs made by nature sound recordist Chris Watson.

The score, developed with Watson and sound designer Chris Timpson of Aurelia Soundworks, follows the arc of a dawn chorus (compressed into 10 minutes rather than 90), shifting from natural to artificial over its course and accompanied by ambient light that mimics the rising dawn.

The installation will be unveiled at Somerset House’s Embankment Galleries as part of the exhibition 24/7: A Wake-Up Call for our Non-Stop World which opens on 31st October. Machine Auguries will also be the first installation at A/D/O by MINI’s new venue in Berlin in spring 2020.

Ginsberg said her interest in AI had grown from her work on synthetic biology adding: “I want to explore how AI researchers are also on a quest to develop artificial life and consider at the same time how humans are neglecting to care for existing lifeforms.” As a growing number of artists have begun to embrace the use of AI, she sees the technology as a tool rather than a substitute for human creativity. “I’m curious how the artist works with AI, rather than submitting to it completely,” she said.

Faculty’s Chief Commercial Officer, Richard Sargeant, said the project demonstrated the potential benefits of technology in art. “The use of AI in art is a fascinating opportunity: in contrast to almost all the uses of AI in industry, this demonstrates the creative, rather than the analytical potential of this technology.”