Neocortex Logo
Coqui shut down in 2024 and its XTTS weights are licensed for non-commercial use only. A commercial-ready alternative for voice cloning in games.

Coqui XTTS Alternative for Commercial Game Voices

ComparisonAI VoiceLocal AI

The problem with Coqui XTTS is not the voices. It is the licence.

Coqui TTS and its XTTS model are open-source voice cloning tools, and many game developers prototyped character voices with them: a few seconds of reference audio, a Python script, and a character speaks. Then the prototype turns into a game you want to sell, and the questions start. Who licenses the model? Who keeps it running? How does it get into Unity?

Can you ship XTTS voices in a commercial game?

Not under the published licence. XTTS-v2 is released under the Coqui Public Model License (CPML), which permits non-commercial use of the model and of its outputs. The toolkit code is under a separate, permissive licence, but that does not extend to the XTTS weights.

Before it closed, Coqui sold commercial licences for XTTS. With the company gone, there is no obvious place to buy one, and the licence on the published weights has not changed. The licence covers outputs as well as the model, so rendering lines offline and shipping only the audio files does not sidestep it. So a cloned voice that sounds right in your prototype can still be one you cannot put in a paid game. (We are not lawyers, and this is not legal advice. Read the CPML yourself.)

Where Coqui stands now

  • The company is gone. Coqui shut down at the start of 2024, and its hosted service went offline.
  • The original repository is dormant. New work happens elsewhere.
  • The code lives on in a fork. The Idiap Research Institute maintains a community fork and publishes the coqui-tts package.
  • Nobody supports it commercially. There is no vendor to call, no service level, and no engine SDK.

What self-hosting XTTS means for a game team

Even with a licence sorted, running XTTS in a shipped game is a project of its own. Here is what your team would build and maintain:

  1. An inference server with a GPU, sized for your peak number of players speaking at once.
  2. Streaming, so speech starts before a full line is synthesised.
  3. An engine client for Unity or Unreal: sending text, receiving audio, playing it back without gaps.
  4. Speech recognition, separately, if players talk back.
  5. A language model and character logic, separately, if NPCs answer rather than read lines.
  6. Voice management: storing reference clips, versioning voices, keeping them consistent across updates.
  7. Upkeep: Python dependencies, CUDA drivers and fork updates, for as long as the game is live.

Each item is doable. Together they are a voice platform, which is a different product from your game.

The costs also behave differently from a game's revenue. A GPU server bills by the hour whether or not anyone is talking, and it has to be sized for the evening peak, not the average. If a streamer sends a wave of players into your game, the voice server is the first thing to fall over.

Self-hosted models vs a managed platform

The choice is less about Coqui and more about two approaches:

Self-hosted open modelManaged character platform
Upfront costEngineering time and GPUsA monthly plan
Ongoing costHosting and maintenancePlan credits
LicenceDepends on the model; XTTS-v2 is non-commercialSet by the plan; Neocortex allows commercial use on Pro and Team
ControlFull control of weights and serversManaged for you; self-hosting on Neocortex is a Custom enterprise arrangement
Engine supportYou write itSDKs included
Offline playPossible, if you ship the modelNeeds a connection, apart from local text chat prototypes

Self-hosting suits research, tools and non-commercial projects. For a commercial game, the licence and the maintenance list above usually decide it.

Coqui XTTS vs Neocortex

Coqui XTTSNeocortex
CompanyShut down; community forkActive, with regular SDK releases
Commercial useModel licence is non-commercialIncluded on Pro and Team
Voice cloningYesFrom a short sample you have the rights to
Voice designNoGenerate voices from a text description
HostingSelf-host on your own GPUManaged; self-hosting as a Custom arrangement
EnginesNone, Python packageUnity, Unreal Engine 5, Godot (beta), Web API
Speech recognitionNoStreaming speech-to-text
Speaking charactersNoFull AI characters with voice in and voice out
Local AIRuns locallyLocal models through Ollama in Unity for text chat prototypes

If local AI matters to your project, our guide to running LLMs locally with Unity covers the Ollama route in detail.

Ship a voice you can sell

Moving an XTTS character takes four short steps. Take the reference clip you cloned from, as long as you have the rights to that recording. Clone it in Neocortex, or describe the voice if you would rather start fresh. Build the character around it in the editor, and replace your Python inference call with the Unity SDK, the Unreal Engine 5 integration, Godot (beta) or the Web API.

Neocortex replaces the whole XTTS stack with one account. Clone a voice from a short recording you have the rights to, or design one from a text description: Pro includes 5 custom voices and Team includes 20, and both plans cover commercial use. There is no GPU to run, since speech is generated and streamed for you. Attach the voice to a character with a personality, knowledge and actions, test it free in the browser editor, and drop it into Unity, Unreal Engine 5 or Godot (beta).

Coqui details were last checked in October 2026. This post is not legal advice; read the CPML yourself before using XTTS.

Frequently Asked Questions

Can Coqui XTTS be used in a commercial game?

The XTTS-v2 weights are under the Coqui Public Model License, which allows non-commercial use only, and with Coqui closed there is nobody to sell a commercial licence. Voices you clone or design in Neocortex can be used commercially on the Pro and Team plans.

Is Coqui TTS still maintained?

The original repository is dormant, while a community fork maintained by the Idiap Research Institute publishes the coqui-tts package. Neocortex is a maintained, managed service with SDKs for Unity, Unreal Engine 5 and Godot (beta).

Does Coqui XTTS have a Unity plugin?

No, XTTS is a Python package you host yourself. Neocortex has a Unity SDK for Unity 6 with six sample scenes, streaming speech in and out, and microphone support in WebGL builds.

Can AI voice cloning run offline in a game?

Running XTTS offline means hosting it on your own GPU under its non-commercial licence. Neocortex voice runs as a cloud service; you can prototype text chat with local models through Ollama in Unity, and self-hosting the platform is a Custom enterprise arrangement.

Written by

Neocortex Team

Date

Oct 17, 2026

Updated

Oct 20, 2026

Read time

5 min

Similar Posts