The vast majority of current artistic activity around AI systems is in the visual domain. However, tools are arriving daily which operate in different domains. One area I find particularly fascinating is speech and music generation.
This work uses the bark text-to-speech model, which I used a few weeks ago to generate various singers for Nietzsche quotes
The song is based around this idea: what if our digital world was being manipulated by a conspiracy of crows?
I asked chatGPT help me come up with some rhyming couplets which would suit this idea, then sent them off to bark to be vocalised. One of the more delightful aspects of bark is that you often get quite unexpected voices and inflections – it reminds me very much of picking a random seed in FFHQ stylegan and getting a completely new face – bark works in a similar way, but with human voices.
I wanted to utilise this pleasing plurality in the track and accompanying video. The people featured are all culled from various US news reports from the 80s and 90s, and AI-lipsynced into the track, to give the impression of rapidly switching channels, only to find a single (rhyming) story on every station.