Becoming L
December 2024 - May 2025
Overview
Becoming L is an ongoing human-AI collaboration and multimedia experience exploring the dialogue between two versions of the same persona: L, an artificially intelligent being representing trans-artist Luca’s former pre-transition self, and their present identity. The installation presents an immersive, multilayered audio piece featuring a conversation between a voice model trained on Luca’s pre-transition voice and their current voice.
My role
I was responsible for training creating the AI voice models using three open-source models and tools. (StyleTTS 2, Coqui’s XTTS-v2, and Chatterbox). Since the source audio constituted intimate personal data, the training was done on offline, local systems instead of using commercial tools.
Team
Luca Lee (Lead Artist)
Key Technologies
Kokoro TTS
StyleTTS 2
ChatterBox
XTTS-v2
Output
This collaboration produced the following dialogue-based audio works, each constructed through iterative exchanges between the synthetic voice model and Luca’s present voice. The source material for L’s voice was assembled from preserved digital fragments: old voice notes, video recordings, and archived audio spanning years prior to transition. These artifacts, never intended as training data, were repurposed into a corpus from which the model could learn the particular rhythms, inflections, and timbre of a voice no longer spoken. The trained model was then used through both text-to-speech synthesis and voice cloning over current recordings, with the resulting outputs carefully curated and edited to shape the final piece.