Light and sound pollution from our 24-hour urban lifestyle affects birds, which are singing earlier, louder, for longer, or at a higher pitch to defend their territory and find mates. Not all birds can adapt: populations are diminishing, and so too the dawn chorus.

In this sound installation, a natural dawn chorus is infiltrated by artificial birdsong. As the room illuminates and the bird calls trained by machine learning get louder, with a call and response set up between natural and artificial birds, we question how the city might sound with changing or disappearing bird populations.

The composition follows the arc of a dawn chorus, compressed into 10 minutes from the natural 90 minutes. The listener experiences the sound of a fictional urban parkland, entering in the dim silvery light of pre-dawn. We start with a lone ‘natural’ redstart, his solo a warbling call. In response, from across the room, we hear an artificial redstart sing back, sampled from an early epoch (an epoch is a complete cycle of training during the machine learning process). A ‘natural’ robin joins the chorus, with a call and response set up between natural and artificial birds, the chorus rises with other species entering, reaching a crescendo 5 minutes in. As the decline starts and the room illuminates to a warm yellow, we realise that the artificial birds, which have gained sophistication in their song, are dominating. The chorus ends with a lone call from an artificial great tit.