Replica Studios Shut Down: An Alternative for Game Voices

You reopen a project you shelved a while ago, press Play, and the guard who used to bark at the player says nothing. The Replica plugin is still in the project. The service behind it is gone.

Replica Studios was one of the first AI voice companies built for games. The Brisbane company offered licensed AI voice actors and plugins for Unity, Unreal and iClone, and in early 2024 it signed an agreement with SAG-AFTRA covering AI voice replicas in games. On June 30, 2025, Replica shut down, citing a tough market for startups and no path to profitability. Its API stopped responding, and users had been told to download their data before the end of June.

If your game still depends on Replica voices, this post is about getting them back, and choosing a setup that will not leave you here again.

Who this is for

You will get the most from this if you are one of these:

• A Unity or Unreal developer whose project still references the Replica plugin.
• A team that exported Replica lines before the closure and now needs new ones in the same voices.
• An iClone or machinima creator who used Replica voices for characters.
• Anyone who tried Replica's Smart NPCs demo and wanted NPCs that actually answer the player.

If you only need to read a fixed script aloud once, plenty of text-to-speech services can do that. The rest of this post is for projects where voices are part of the game itself.

What Replica Studios offered, and what stopped

Replica's product was a library of AI voice actors. You picked a voice, typed or imported lines, adjusted delivery and exported audio, either in its web app or through engine plugins. It also showed Smart NPCs, an Unreal demo in which players could speak to characters in a city and hear generated replies.

When the company closed, all of that went with it: the voice library, the generation API, the plugins, and the accounts that held your projects. Audio files you exported still work, because they are just files. Anything you planned to generate later does not.

Two kinds of Replica projects

Scripted voice-over lines

Cutscenes, barks, tutorial prompts and announcer lines are written in advance and rendered to audio. For these, the loss is the ability to add or change lines in a voice the player already knows. One new quest means a voice that does not match.

Speaking NPCs

Here the character answers whatever the player says, out loud, at runtime. That needs more than a voice: speech recognition, a language model that stays in character, and a way for the reply to affect the game. Replica's demo pointed at this. Neocortex is built for it.

Rebuilding your voice cast

Start with the voices players will notice first, usually your protagonist's companions and the NPCs in the opening hour.

First, create a free Neocortex account. Then, for each character, choose how to recreate the voice. If you have the rights to a recording of the original performer, clone the voice from a short sample. If the Replica voice was a stock AI actor you never had rights to outside Replica, design a new voice instead: describe it in a sentence, such as "gravelly older sailor, slow and amused", and pick from the generated options.

Next, build the character around the voice. In the visual editor, the Agent, Personality, Emotion and Knowledge nodes define who it is, and Intents and Actions define what it can do in your game. Talk to it in the browser until it sounds right. In-editor testing never uses credits.

Finally, connect it to your NPC. Install the Unity SDK (Unity 6, six sample scenes) or the Unreal Engine 5 SDK, or call the Web API from another engine. Your game sends nearby objects and recent events, and each reply comes back with an emotion tag and any actions the character decided to take. Microphone input works in Unity WebGL builds as well.

Keep the scripted lines you exported where they still fit, and let the rebuilt voice carry everything the character says at runtime, so new dialogue stays close to the voice players already know. Pro includes 5 custom voices and Team includes 20.

Choosing where your voices live

Replica's closure was not unique. Our timeline of AI character vendor shutdowns covers several more, and each one left teams with a similar mess. When you rebuild, it helps to weigh the approaches, not only the vendors:

One cloud voice vendor for everything. Quick to start, with no servers to run. The trade-off is that the voices, the generation and your project data all live in one place, which is exactly what failed here.

Self-hosted open models. You keep the weights and run them yourself. The trade-off is GPU hosting, an engine integration you write, and model licences that often rule out commercial use.

A character platform with an exit path. Voices, characters and engine SDKs in one service, plus ways to reduce dependence on it. This is where Neocortex sits: you can prototype character conversations with local models through Ollama in Unity, and self-hosting the platform is available as a Custom enterprise arrangement for studios that need it.

The bottom line

Replica gave games licensed AI voices, and its closure took the voices and the plugins with it. Neocortex gives you the voice back, through cloning or voice design, and adds what Replica only demoed: characters that hear the player, answer in that voice and act in your game, in Unity, Unreal Engine 5, Godot (beta) or any engine through the Web API.

Details about Replica Studios come from public reporting on its closure, checked October 2026.

Frequently Asked Questions