Do

Description

VOXAIPE is a typographic experiment where the sound of a person’s voice directly generates a changing typeface in real time. Instead of focusing on what is being said, the project focuses on how the voice sounds, its pitch, timbre, rhythm, and other vocal qualities. The work shifts from traditional typography toward something like phonography. It also explores a bigger question: when AI translates us, are we seeing an honest portrait, or just another performance or simulation?

Process

VOXAIPE began as a technical question: what happens when the sound of a voice designs a typeface directly? But it quickly became a conceptual shift in my practice. In earlier work, I argued for more control over the AI systems that interpret us. With VOXAIPE, I deliberately moved in the opposite direction: I let the system form its own logic and confronted what it means to be translated without being able to steer the translation. That decision comes from lived experience of being interpreted before being understood, by people, institutions, and increasingly by algorithms. The project grew from that history into a larger question: is agency only about control, or can it also be the conscious choice to give up control in order to create a different kind of encounter? VOXAIPE reframes agency as resonance, building a closed-loop system where voice becomes a renewable input that returns as feedback for self-reflection.

Tools

VOXAIPE is built as a real-time audio-to-typography pipeline. First, live voice input is captured as the raw material. An acoustic analysis model, wav2vec, then deconstructs the audio into a high-dimensional vector of vocal characterisitics such as pitch, timbre, and rhythm. This vector is fed into DeepFont, a neural font architecture that generates a variable font which continuously morphs on screen as the person speaks. A crucial aspect of the process is that no fixed design rules were programmed (for example, there is no instruction like “high pitch equals bold”). Instead, the AI develops its own internal logic for translating sound into shape. Letters may expand into wide, heavy forms, soften into rounded terminals, or suddenly condense into sharp, blade-like serifs depending on the vocal qualities in the moment. Because this live transformation is fleeting, the project is also captured in “DECODED,” a 228-page artist book that documents an ongoing monologue between the artist and the machine. The book contains no words, focusing instead on form and time to preserve the visual trace of that real-time exchange.