How does voice acting for games work?
Very differently from film — actors typically record alone, out of order, without seeing the game, which is why direction and documentation matter enormously and why performances sometimes feel disconnected.
The constraints:
Non-linear scripts. A game's dialogue branches, so an actor records every variant of a conversation without knowing which the player will hear.
Out of context. Lines are frequently recorded years before the scene exists visually, from a spreadsheet rather than a screenplay — sometimes with no plot summary at all, particularly on projects with secrecy requirements.
Alone. Co-actors are rarely present, so reactions are performed against nothing. Some studios now record scene partners together, which produces noticeably better results and costs more.
Enormous volume. A large role can run to tens of thousands of lines, including hundreds of efforts — grunts, jumps, injuries, deaths — which are physically demanding and a documented cause of vocal strain.
Technical requirements — consistent distance and level across sessions months apart, and delivery that leaves room for in-engine processing.
How performance capture differs. Face, body and voice recorded simultaneously on a volume, with actors in motion-capture suits. It produces far more coherent performances and is considerably more expensive, so most games mix both.
Why localisation multiplies everything. Each language repeats the whole process, with the added constraint of fitting existing animation timings — which is why some localised versions sound clipped or oddly paced.
The industry issues that have driven disputes: vocal stress from effort-heavy sessions; residual payments, where a game sells for years and actors are paid once; secrecy preventing informed consent about roles; and synthetic voice replication, which became the central issue in recent industrial action — specifically whether a performance can be used to train a model that then generates new lines without further consent or payment.
What good practice looks like: context provided, session lengths limited, efforts scheduled at the end, and a director present rather than an engineer alone.