Singing and weeping and laughing to teach an algorithm (study I) 2023

Sounds emitted by human bodies can be training data for algorithms used in voice interfaces for digital assistants like Alexa or Siri. The data is collected by human annotators who verify the sounds found online, mainly on YouTube videos. The extracted sounds are then categorized and labelled manually. A vast human sounds repository is created to feed the algorithmic audiences that grow bigger year by year. 

Listening to and learning from the sounds that our bodies produce could establish a very intimate connection with a machine. How will our interaction and relations with machines evolve through sound? What kind of infrastructures are behind these seemingly fully automated systems? 

This video study is part of an ongoing research on voice interfaces and digital assistants initiated in 2021 with the video installation Not allowed for algorithmic audiences. The installation was commissioned for the ArtScience Residency by Ars Electronica, Art Collection Telekom, along with Wild Alchemy Lab.

‍

Images by Kyriaki Goni