Arc Raiders AI Voice: What Embark Studios Uses, and Why

When Arc Raiders launched in October 2025, players noticed that some character and call-out lines sounded synthetic. Embark Studios confirmed it had used AI text-to-speech for part of the game’s dialogue, and the decision drew criticism from players, from voice actors and from parts of the games press.

The criticism was not that the game sounded bad. It was about principle: whether a big commercial game should use generated voices in place of paid performances, and whether the actors whose voices trained the models had fair terms.

The game was a commercial hit regardless, which kept the question in the spotlight. Searches for “Embark Studios Arc Raiders AI voice” jumped again on November 3, 2026 after Embark published a video in which its audio team explained its approach.

Embark has given a fairly clear picture in interviews, though it has not published a line-by-line list.

CEO Patrick Söderlund told Engadget that “a lot” of the voice lines are recorded by human actors, and that there are fewer AI-generated lines than at launch. Windows Central and others reported that the AI voices that remain are trained on, and licensed from, actors Embark hired and paid.

The new Embark video focuses on pings, the quick spoken call-outs a player triggers to mark loot, enemies or places for teammates. Audio director Bence Pajor explained that the list of things you can ping keeps growing, even late in development and after launch.

Every new item, enemy or location needs a line, often in several voices and languages. Recording each one means booking studio time with each actor for each update. A licensed AI voice lets the team add a call-out the same day a new item ships.

Kotaku’s coverage was sceptical, quoting the video’s point that actors “would just go crazy if they had to read all of that.” Supporters say the practical case is real as long as actors consent and are paid. Critics say it sets a precedent that will shrink paid voice work across the industry.

Arc Raiders has become a test case because it is popular and because Embark is open about what it does. The pattern it shows is likely to spread:

• Hybrid audio. Human performances for story and emotion, AI for large volumes of short, frequently updated lines.
• Consent and pay. The defensible version uses voices licensed from actors who agreed and were paid, not scraped audio.
• Disclosure. Players increasingly expect studios to say where generative AI is used. Steam already asks developers to disclose it.

For voice actors and unions, the open question is how licensing terms and pay should work when one recording session can produce unlimited future lines.

Every AI-generated line runs on a GPU somewhere. For one game the load is small, but voice, dialogue and character AI across the whole games industry, plus every other business using speech models, add up to steady demand for inference capacity in data centers.

That demand is one reason smaller AI data centers and converted Bitcoin mines are being built across Canada and the US. They serve inference workloads that need low latency and reasonable power prices more than they need the largest training clusters.

If you are pricing that kind of facility, the AI Build Configurator shows capital cost, power cost and payback for any Canadian province or US state in seconds. The demo is free.