A voice cloning model and a Telegram bot
A voice cloning model and a Telegram bot built to preserve the sonic identity of a film assistant director while he undergoes a laryngectomy process.

Javier Soto is an assistant director with a career built on high-level international film sets. He has worked with J.A. Bayona on Society of the Snow, with Elia Suleiman on 7 Days in Havana, and alongside directors like Oliver Stone and Jonathan Glazer. The film set is a space where the voice literally rules. It coordinates teams, transmits decisions, sustains the rhythm of an entire production. When Javier received a diagnosis of laryngeal cancer, he asked us for something that rarely becomes a technical project: to clone his voice before it disappeared. This short documentary, developed in collaboration with Dutch visual artist Jan Rothuizen, starts from that request to build research into what it means to preserve the human voice with artificial intelligence. Jan is a visual artist, illustrator, and filmmaker. His project Soft Atlas has been mapping cities since 2009 through hand-drawn maps that mix observation, conversation, and narration. He has brought this method to formats such as the interactive documentary Refugee Republic and the VR film Drawing Room. His latest piece, Tracing Colombia, was presented at IDFA in 2025. Rothuizen works at the boundaries between drawing, testimony, and technology.

The developed system combines two models. An LLM generates the text of the responses based on a prompt system that reproduces Javier’s conversational profile, his way of speaking, his references, his way of moving through the film world. And a trained audio model converts that text into audio using a voice synthesis model trained with Javier’s original recordings. The result is a Telegram bot with two modes of operation: Javier writes or speaks with his new voice and receives a voice response. The technical architecture is built on Node.js, with persistent conversation management, a per-user authorization system, and remote administration.
The bot allows Javier to communicate through his cloned voice in situations where the tracheoesophageal valve he uses after the laryngectomy limits spontaneous communication. A functional, discreet tool, integrated into an application that is already part of his daily life.

A laryngectomy is not just a medical intervention. It is a transformation of identity. The voice is the most immediate instrument of presence, the channel through which a person becomes recognizable. This project is part of the «AI for Good» tradition, where language and voice synthesis models stop being productivity tools to become systems of personal continuity. Technology does not return the lost voice, but it builds a bridge between what was and what can continue to be.

This project always reminds us of The Little Mermaid. In Hans Christian Andersen’s tale, Ariel is a mermaid who gives her voice to a sea witch in exchange for human legs. She wants to live on land, be near the man she loves, belong to a world that is not her own. The price she pays is the most intimate one imaginable. Without a voice, Ariel exists but cannot make herself recognized. She loses the instrument through which she was herself. The story ends in tragedy because the voice does not return, and without it, identity dissolves. There is something in that image that takes hold strongly when working on a project like Javier’s. The difference is that here, technology makes possible what the fairy tale denies. The voice is preserved, cloned, kept active before the body loses it. The resistance to yielding that territory, which in Andersen is impossible to sustain, finds a form of response in artificial intelligence. This tension between identity, perception, and technology crosses several Artefacto projects. Yet, The Faces starts from a symmetrical and complementary question. If in Javier’s project the technological system preserves a voice so a person remains recognizable, Yet, The Faces investigates what happens when recognition fails from within. Two projects working the same territory from different angles. In one, technology extends identity beyond the limit of the body. In the other, technology reveals that recognition is always a fragile process, both for machines and for people.
Code and documentation: github.com/jcaballeroramos/javier-soto-bot-telegram