Artificial Félix Dances the Transversal Intellect

The recording is part of a concert event and a live radio broadcast by “radio ansteckende symbiosen” on Freies Radio Lora in Zurich. The concert took place on July 21, 2026, at the Kunstraum Walcheturm in Zurich.

There are two textual layers.

The first consists of three verses from the song, spoken by an AI voice based on the open-weight model Qwen3-TTS, which was trained using a video interview excerpt (in French) by Félix Guattari (who conceputalized "transversality" in the 1960s). The generated speech was subsequently adapted to match the rhythm of the music.

- Dance the transversal intellect,

- more-than-individual,

- collective intelligence

The second layer of text consists of three text fragments sung by an AI voice, the content of which relates to the production of the piece. It was post-processed using an OpenWeight model Spleeter from Deezer to isolate the vocal track. Afterward, the melody and pitch were recomposed using Melodyne to fit the song

- Hacke die Webseite und entlocke ihr die m4a Datei. Bring deine CPUs zum glühen. [Hack the website and extract the m4a file from it. Make your CPUs run hot.]
- Nimm die Abfälle der KI Prozesse, die Demos der Musik-KIs. Stoppe dort, wo ein Abo fällig wird. [Take the waste products of AI processes—the demos from music AIs. Stop where a subscription is required.]
- Entreiß der mainstream Langeweile ihre Stimmen. Re-tune the voices. [Snatch the voices from mainstream boredom. Re-tune the voices.]

The third layer is a sound layer that uses the recording of the lyrics from the first verse and the chorus, which were sung by an AI (demo mode of the AI at suno.com, hacked). It has also been post-processed with Spleeter from Deezer and re-tuned in pitch and time.

It is then played live using software developed in Pure Data, which enables montages, time stretches, and compressions via a MIDI controller. This software also has an accompaniment mode that attempts to play along with the improvised montages and time distortions. The accompaniment is derived from the Rhodes piano part of the song and converted into a MIDI track using open-source AI, which the Pure Data software then utilizes. The MIDI data is sent to Ardour. There, the Sfizz plugin plays the incoming MIDI notes using a Fender Rhodes multisample. The sounds are then sent to an open-source AI model of a Fender Twin Reverb (NAM model: Fender Vintage Reissue 65 Twin Reverb), which was trained using two microphones (Neumann U 67, Royer R-121).

The video file will be released on 25 Nov. 2026.