CD Projekt uses AI to rejuvenate Geralt's voice after his Polish actor's stroke: a consent-based use that sets the bar

🕒 Published on Zendoric: November 12, 2026 · 00:05
✨ AI-generated · how it's made
For The Witcher 3 expansion "Songs of the Past," CD Projekt Red will use Respeecher so that the voice of Jacek Rozenek, who was affected by a stroke, sounds as it did in 2014. It's a case of AI as a consensual vocal prosthesis in an industry that has been badly burned.
By IGN · November 11, 2026.
CD Projekt Red (CDPR) will use an AI voice tool to alter the voice of the Polish actor who dubs Geralt of Rivia in The Witcher 3's “Songs of the Past” expansion, more than 11 years after the game's release. According to IGN, citing an interview with Game File, the tool is Respeecher.
The actor, Jacek Rozenek, suffered a stroke in 2019 that left him unable to speak for a time. He regained his voice after intensive training and had already taken part in a couple of quests for the 2022 next-gen update, but he told Game File that recording left him “exhausted.” His voice today is noticeably deeper than in 2014, when he recorded most of Geralt's lines. The new lines would sound odd next to the old ones, which are set in the same period of the game.
The solution was to train Respeecher's model on Rozenek's own original lines. It isn't a quick filter. According to Respeecher synthetic voice artist Dmytro Vasylets, getting the result took “dozens of hours” of model training, adjustments with the CDPR team and dealing with “AI artifacts that never fully go away” (audio flaws typical of synthesis). The performance is still the actor's; the model changes the timbre.
Rozenek gave his approval. He said he has “great trust” in the game's creators and feels “very good” about the dynamics and the performance aspects. CDPR, through its localization producer Mikołaj Szwed, says it will not alter any actor's voice without their express permission. The company also maintains that it won't be necessary in The Witcher 4, where Rozenek is expected to return, because an older Geralt can have a deeper voice.
Our take: this case is useful because it separates two things the debate over AI in video games tends to conflate. One is replacing a performer with a synthetic replica to cut costs. The other is using AI as a prosthesis that lets an artist keep working despite a health condition. Here the actor is still performing, has consented and is the source of the training data. It is the most defensible use we have seen, and it is worth setting as a benchmark: explicit consent, the actor's own data, the actor at the center and public transparency.
That doesn't make the problems go away. The industry as a whole remains very tense about AI: IGN notes that the Indie Game Awards stripped Clair Obscur: Expedition 33 of its Game of the Year award over placeholder textures generated with AI, and that AI-generated art was found in Crimson Desert, since removed in a patch. The short-term risk is that a good precedent becomes cover for less clean uses. CDPR's assurances are company statements, not verifiable contracts, and actors' unions will keep demanding written rules.
In the long run, this is the friendly face of the technology we champion: tools that give back abilities to people who lost them to illness or age. If medicine and AI reduce those losses, these prostheses will become ever rarer and more valuable. In the meantime, the yardstick is simple: who consents, who controls the model and who gets paid for it?
🔗 Related on Zendoric
- Hasbro Asks Child Voice Actors to Hand Their Voices to AI: A Test Case for Consent in the Synthetic Era · 2026-06-28
- Hasbro Asking Young Peppa Pig Actors to License Their Voices to AI Tests Where Consent Should Start · 2026-06-27
- Trademarking the voice: the legal play with which artists get ahead of deepfakes · 2026-06-27


