CD Projekt Red is using Respeecher AI to shape newly recorded Polish dialogue for Geralt in The Witcher 3: Songs of the Past, aiming to bring Jacek Rozenek’s voice closer to his earlier recordings after changes following his 2019 stroke.

Rozenek approved the approach, saying he trusts the game’s creators and is comfortable with the acting and vocal dynamics; Respeecher says the work required dozens of hours of model training and close collaboration, particularly to preserve his emotional range while maintaining audio quality.

The studio says every line is reviewed through a human-in-the-loop process, and describes the project as “a special case,” not necessarily a broader shift toward synthetic voices across its games.

The approach lets the studio retain Rozenek’s performance rather than recast Geralt or accept a substantially different voice, though the article asks whether hiring and paying a human replacement might have been preferable; finding one and involving Rozenek in training them could also be difficult and uncomfortable.

The adjustment is limited to the Polish performance: Doug Cockle’s English-language Geralt dialogue will not receive the same AI treatment.

CD Projekt Red has not disclosed the expansion’s exact release date, price, platforms, number of new lines or amount of training data, and similar processing in other language versions is unconfirmed.

Songs of the Past is expected in 2027, 12 years after The Witcher 3 launched, and will introduce the region of Letten with a story centered on Dandelion and his family.

Although the expansion is listed for December 31, 2027, the article says that date appears to be a placeholder rather than a confirmed launch date; the previous major expansion, Blood and Wine, arrived in May 2016.

The expansion will not be available on PS4 or Xbox One, despite The Witcher 3 being playable on those consoles.

The case stands apart from many concerns about AI voice replication because Rozenek is alive and has personally consented, amid broader debate over performers’ control of their voices and likenesses, including issues raised during the 2024 SAG-AFTRA video game strike.

Respeecher says it requires explicit permission and a signed agreement for voice replication, though its promotional materials also market the technology for creating game character voices at scale.

The project may offer an example of a consent-first workflow, but it does not set a binding industry precedent; its significance will depend on transparency, compensation, rights terms and how the finished audio sounds to players.