Neiroha VoxCPM2

This page covers the Neiroha VoxCPM2 local backend. Extract the portable Release package, or place the source repository anywhere; <backend-root> refers to that directory.

It provides OpenAI-compatible routes, native /api/voxcpm routes, a voice registry, Neiroha Gradio Admin, and optional official WebUI. Startup behavior is controlled by configs/server.toml.

The backend Admin loads models, manages the voice registry, runs tests, and displays logs.

Capability Summary

Dimension	Current Notes
Recommended version	VoxCPM2 is the current official deployment version, 2B parameters, 48 kHz output.
Languages	Official list includes 30 languages: Arabic, Burmese, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Indonesian, Italian, Japanese, Khmer, Korean, Lao, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Tagalog, Thai, Turkish, Vietnamese.
Dialects	Official list includes 9 Chinese dialects: Sichuanese, Cantonese, Wu, Northeastern, Henan, Shaanxi, Shandong, Tianjin, Minnan.
Cross-language output	Supports multilingual synthesis and cross-language reference-audio cloning.
Text voice design	voxcpm2-design needs no reference audio; put age, gender, tone, emotion, and speed in natural-language text.
Controllable clone	voxcpm2-clone uses reference_audio and does not need prompt text.
Ultimate clone	voxcpm2-ultimate-clone needs prompt_audio and exact prompt_text; do not rely on bracket style control in this mode.
Official speed reference	Official PyTorch reports RTF around 0.30 on RTX 4090; Nano-vLLM / vLLM-Omni acceleration is around 0.13.
Boundaries	Very short text may sound weak. Long text can speed up, produce noise, fail to stop, or OOM; split by sentence for production.

Default Addresses

Service	Default Address	Purpose
FastAPI	http://127.0.0.1:8000	Neiroha provider connects here.
Admin	http://127.0.0.1:7860	Manage voices, model presets, tests, and logs.

Install

Recommended portable flow. The current package is built for NVIDIA GPU / CUDA environments and mainly targets RTX 30 / 40 / 50 series users:

1. Open Neiroha-VoxCPM V1.0.0 Release.
2. Download all V1.0.0 split archives: Neiroha-VoxCPM-portable.7z.001, .002, .003, .004.
3. If GitHub downloads are unstable, use the Baidu Netdisk mirror from the Release body.
4. Put all four files in the same directory and extract from .001 with 7-Zip.
5. Run start_portable.bat.

Source or development environment:

pixi install
pixi run install

Optional ASR model for automatic prompt text transcription in ultimate clone:

pixi run install-asr

ASR is disabled by default. ultimate_clone normally needs manual prompt text.

Start

Portable Release:

.\start_portable.bat

Source environment:

pixi run serve

Common tasks:

Command	Purpose
pixi run serve	Start according to configs/server.toml [startup].surface, default API + Admin
pixi run api	Start FastAPI only
pixi run admin	Start Neiroha Admin and connect to an existing FastAPI
pixi run smoke	Check /health, /v1/models, /v1/audio/voices, and capabilities
pixi run test	Run backend tests
pixi run launcher-help	Show launcher arguments

Connect Neiroha

1. Open Providers.
2. Create a provider, adapter type VoxCPM2 Native.
3. Set Base URL to http://127.0.0.1:8000.
4. Leave API Key empty if local auth is disabled.
5. Click Fetch All.
6. Confirm voxcpm2-design, voxcpm2-clone, and voxcpm2-ultimate-clone.
7. Enable the provider and click Health Check.

Android emulator host URL:

http://10.0.2.2:8000

Character Setup

Goal	Setting
Text voice design	Select voxcpm2-design; write natural-language voice description at the beginning of the text.
Reference clone	Select voxcpm2-clone; provide reference audio, no prompt text required.
High-fidelity clone	Select voxcpm2-ultimate-clone; provide prompt audio and prompt text.
Reusable local speaker	Register a voice in Admin or /api/voxcpm/voices, then select it in Neiroha.

VoxCPM2 recommends natural-language style prompts, for example:

(A young woman, gentle and sweet voice)Hello, welcome to VoxCPM2.

Do not depend on undocumented square-bracket tokens.

API Prefix

OpenAI-compatible routes:

Method	Path	Purpose
GET	/health	Health check
GET	/v1/models	List voice sets
GET	/v1/audio/voices	List voice profiles
GET	/v1/audio/speakers	Speaker-list compatibility
POST	/v1/audio/speech	OpenAI-compatible synthesis

Native prefix is /api/voxcpm. Legacy /voxcpm/* routes remain for compatibility; new integrations should use the standard prefix.

Sources

• OpenBMB/VoxCPM README
• VoxCPM2 model notes
• VoxCPM2 Usage Guide