This app begins with a single whisper but, as the device is held still, more and more whispers join to form a chorus of indiscernible language intended as a more responsive, human alternative to a white noise machine.
The whispers in this video are generated from an AI model of my own voice, intended to generate whispers that are not only inaudible but have no meaning at all behind them – that is, they take only the aesthetic qualities of a whisper without any of the meaning behind the sounds it hears. The app generates snippets of whispering on the fly, with no audio data stored on the device.
I used WaveGAN, which generates audio in snippets, so I use layers of audio to give the illusion of constant speech. Groups of layers are faded in and out according to the motion of the phone – when the phone is moved, each group fades out until only one is left. The other groups fade back in at different rates once the phone is held still, either in the hand or on a surface – the motion indexing algorithm is tolerant of small movements, so it’s possible to move the phone slowly and not lose the full chorus of whispers.