Jasia Reichardt Lecture 2027: Atau Tanaka

Title: Jasia Reichardt Lecture 2027: Atau Tanaka
Date: 20th January 2027, 18:00
Location: BCS Moorgate, London / Zoom

Book on Eventbrite Add to calendar

Atau Tanaka introduces variational auto-encoders for audio and explores timbre transfer using RAVE. He discusses two contemporary musical works trained on recordings of their own performers, considering data provenance, ethics and embodied interaction as musicians shape AI-generated sound live in performance.

AI Generated Summary

About the Event

We are very excited to announce that the Jasia Reichardt Lecture 2027 will be given by Prof Atau Tanaka of Goldsmiths, University of London and the Bristol Interaction Group. In his talk he will introduce the technique of variational auto-encoders, thinking of them as encoders and decoders.

Seen in this light, they are familiar to musicians and audio engineers, who work with codecs and domain transformations. Tanaka will present RAVE, a variational auto-encoder for audio developed by Caillon and Esling at IRCAM, distinguishing its analysis of the audio signal from symbolic music systems that treat musical representations. He will show how a variational auto-encoder applied to sound performs timbre transfer, transferring the sound of one instrument onto another, analogous to style transfer in visual media.

He will then show how RAVE was used in the composition and performance of two works of contemporary classical music: Transfert for percussion, EMG and RAVE, and Burning Meditation for voice and AI. In both cases, data provenance directly addresses questions about the ethics of the training dataset: the models were trained on audio data collected from the musicians who would go on to perform the works.

Body-based interaction via the muscle electromyogram enables performers to modulate the latent space of the AI sounds live, allowing their gestures to influence the variational auto-encoder’s output. The resulting works are dialogues between virtuoso acoustic instrumental and vocal performance and an alien-like electronic avatar arising from models of the musicians themselves. Tanaka will present an end-to-end production cycle, from data capture and training through composition to live performance, putting the human performer into real-time interaction with the variational auto-encoder.

This year's lecture has been made possible with funding from the Westminster Councillors Ward Budget Programme.

Supported by City of Westminster

About the Speaker

Atau Tanaka

Speaker

Atau Tanaka

Prof Atau Tanaka conducts research in embodied musical interaction. He studied electronic music with Ivan Tcherepnin at Harvard University. There he met John Cage during his Norton Lectures. Tanaka went on to carry out his doctorate with John Chowning and Max Mathews at CCRMA Stanford. Atau has carried out research at IRCAM Centre Pompidou, Apple France, and Sony Computer Science Laboratory (CSL) Paris in areas of interactive music, immersive media, network performance, and human-data interaction. His work has been presented internationally at Ars Electronica, ZKM, WOMAD festival, Sónar Barcelona, SFMOMA and Musikinstrumenten-Museum Berlin. It has been reported in the press on BBC Radio 3, BBC World Service Click!, New Scientist, The Wire, and Vice. He has been Edgar Varèse guest professor at TU Berlin, conducts research with the Bristol Interaction Group, and MSH Paris Nord. He has held academic positions at Keio University, Le Fresnoy, and Newcastle University. He is professor at Goldsmiths University of London, co-director of the Centre for Sound, Technology & Culture and is editor of the Sonics Series on Goldsmiths Press.