CD Projekt RED will use AI-based voice synthesis to maintain the original vocal consistency of Geralt of Rivia in the upcoming expansion, Songs of the Past.
The studio is employing Respeecher technology to blend new dialogue with thousands of existing recordings from 2014, ensuring the protagonist’s voice remains uniform throughout the game.
The development team faced a major obstacle as new lines recorded today would naturally clash with audio assets created over a decade ago. To mitigate this, CD Projekt RED engaged Respeecher, a company previously utilized for the Phantom Liberty expansion in Cyberpunk 2077, as reported by Game File. This technology does not replace the actor, Jacek Rozenek, who remains the primary voice performer.
Rozenek records all new lines personally, and the software subsequently modifies the audio output to match the timbre and characteristics of his 2014 performance. “Jde o to, aby se tento hlas co nejvíce přiblížil mému hlasu z doby před dvanácti lety,” Rozenek told Game File. The actor confirmed he supports this process and maintains full confidence in the studio’s technical direction.
This implementation is restricted exclusively to the Polish language version of the game. English-language dialogue, voiced by Doug Cockle, will not undergo similar AI-driven adjustments, as noted by the studio. Mikołaj Szwed, the localization producer at CD Projekt RED, described the project as an ethical application of artificial intelligence, performed entirely with the explicit consent of the original performer.
The approach to AI in the Witcher franchise varies significantly across different community initiatives. While the studio uses high-end synthesis for professional polish, independent projects have also emerged to bring localized audio to the base game. Ondřej Šitler recently coordinated a fan-led effort to add Czech dubbing to The Witcher 3, as reported by CzechCrunch.
This community project utilized a hybrid approach to generate over 62,000 lines of dialogue. “Nicméně to není anonymní syntéza. Dobrovolník nahraje vzorek, z něj vznikne hlasový model a ten pak namluví repliky jeho postavy,” Šitler told CzechCrunch. The team relied on volunteers to record samples, which were then processed to create character-specific models, with the stated goal of eventually replacing machine-generated voices with live human performances.
The following table highlights the use of AI for voice localization in the Witcher ecosystem, as reported by Game File and CzechCrunch.
| Project | Primary Goal | AI Role | Actor Involvement |
|---|---|---|---|
| Songs of the Past (Official) | Vocal consistency | Post-processing synthesis | Active recording |
| Czech Dubbing (Community) | Full localization | Model-based synthesis | Volunteer sampling |
For instance, the team had to manually adjust the script when the software struggled with how to write certain names and words, as reported by CzechCrunch. CD Projekt RED has not announced a specific release date for the Songs of the Past expansion.
