VoxCPM2

VoxCPM2 is an open-source TTS model from OpenBMB with voice cloning support. OpenMAIC includes an adapter, so you only need to run VoxCPM on your own machine and connect it.

When to use VoxCPM2

• You want stable, free TTS with no per-character charges.
• You want voice cloning, with each Agent using a short sample to generate its own voice.
• You deploy locally or in an offline environment.

If you only need one default voice, the built-in Doubao or OpenAI-compatible provider is simpler. See Configuration → TTS providers.

1. Run a VoxCPM backend

OpenMAIC supports three deployment modes. All three use the same OpenMAIC adapter; select the corresponding backend type in Settings.

See the VoxCPM repository for startup instructions for each backend. A typical local quick start:

# vLLM-Omni example
pip install vllm
python -m vllm_omni.server --model openbmb/VoxCPM2 --port 8000
# endpoint at http://localhost:8000/v1

2. Point OpenMAIC at it

Choose one of the following methods.

A. Per-user (Settings UI, no server change)

Open Settings → Text to speech → VoxCPM2, choose the backend type, and paste the Base URL. The Request URL preview below shows the actual address OpenMAIC will call.

This is suitable for personal testing and per-browser overrides, and does not affect other users.

B. Server default (environment variable, shared by everyone)

Add the following line to .env.local or the YAML configuration. No API key is required.

TTS_VOXCPM_BASE_URL=http://localhost:8000/v1

Server configuration is sent to users. If the provider is server-managed, clients cannot override these settings.

3. Voice management

VoxCPM2 has three voice modes under Settings → Text to speech → VoxCPM2 → VoxCPM voices.

Auto Voice (default)

During synthesis, OpenMAIC dynamically generates a voice prompt from each Agent's persona. No configuration is required; this is what you get by doing nothing.

Prompt voice

Describe a voice in natural language. Once generated, it can be reused and assigned to any Agent.

Example: “A warm female teacher voice, calm and encouraging, mid-range, with clear articulation.”

Clone voice

Upload a short reference audio clip (≤ 60 seconds, ≤ 10 MB), or record one directly in the browser. The audio is stored in IndexedDB and sent to the backend with each synthesis request.

Troubleshooting