Neiroha GPT-SoVITS

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

It provides FastAPI, Gradio Admin, TOML model presets, TOML voice sets, an OpenAI-compatible TTS API, and native /api/gpt-sovits routes. Startup is configuration-driven through configs/server.toml.

The backend Admin shows API state, loads presets, manages voices, downloads assets, and displays logs.

Capability Summary

Dimension	Current Notes
Recommended preset	Default model preset is v2proplus-clone, using GPT-SoVITS v2ProPlus SoVITS weights.
Languages	Official cross-lingual inference range: Chinese, English, Japanese, Korean, Cantonese. Neiroha uses zh / en / ja / ko / yue.
Cross-language output	Supported when target language stays within the supported range.
Dialects	Officially explicit dialect support is Cantonese. Other Mandarin regional dialects are better handled by separate training or CosyVoice3 / VoxCPM2.
Clone prompt text	Required. Clone needs reference audio, matching reference text, prompt language, and text language.
Reference audio	Official zero-shot examples use around 5 seconds. The local backend recommends clean 3 to 10 second clips.
Official speed reference	Official README reports v2ProPlus RTF around 0.028 on RTX 4060Ti, 0.014 on RTX 4090, and 0.526 on M4 CPU.
Boundaries	Split long text. Reference text/audio mismatch increases missing words, repetition, and articulation risk.

Default Addresses

Service	Default Address	Purpose
FastAPI	http://127.0.0.1:9880	Neiroha provider connects here.
Admin	http://127.0.0.1:7860	Manage presets, voices, downloads, and logs.

If a port is occupied, the launcher chooses a free random port and writes it to the terminal and runtime/logs/backend.log.

Install

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

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

Source or development environment:

pixi install
pixi run install
pixi run install-sample-voice

These commands initialize the GPT-SoVITS submodule, install upstream dependencies, download base pretrained assets, and install one sample voice. Existing .ckpt / .pth weights can be registered in Admin Model Presets and clone configuration.

Start

Portable Release:

.\start_portable.bat serve

Source environment:

pixi run serve

Common Pixi 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 Gradio Admin only and connect to an existing FastAPI
pixi run smoke	Check /health, /v1/models, /v1/audio/voices
pixi run test	Run backend tests
pixi run launcher-help	Show launcher arguments

admin mode starts Gradio Admin only. To run API and Admin together, keep [startup].surface = "both" and run pixi run serve.

Connect Neiroha

1. Open Providers.
2. Create a provider, adapter type GPT-SoVITS.
3. Set Base URL to http://127.0.0.1:9880, or the actual address printed in logs.
4. Leave API Key empty if local auth is disabled.
5. Click Fetch All.
6. Confirm default voice set and genshin-keqing voice are visible.
7. Enable the provider and click Health Check.

Android emulator host URL:

http://10.0.2.2:9880

Character Setup

Goal	Setting
Default sample voice	Preset / trained voice mode, voice genshin-keqing
Custom trained weights	Add .ckpt / .pth in Admin Model Presets, then create a voice
Reference-audio clone	Provide reference audio, reference text, reference language, and target text language
Multiple voice sets	Create a voice set and expose desired voices to Neiroha

v2ProPlus clone is sensitive to reference length. Use clean 3 to 10 second clips without background music.

Native API Prefix

Stable OpenAI-compatible routes:

Method	Path	Purpose
GET	/health	Health check
GET	/v1/models	List voice sets
GET	/v1/audio/voices	List voice profiles
POST	/v1/audio/speech	Synthesize with registered voice

Native prefix is /api/gpt-sovits. Legacy /gpt-sovits/* and /tts remain for compatibility, but new integrations should prefer the standard prefix.

Logs and Output

• Output audio: runtime/outputs/
• Current log: runtime/logs/backend.log
• Previous log: runtime/logs/backend.previous.log
• Download logs: runtime/logs/admin-download.out.log and runtime/logs/admin-download.err.log

Sources

• RVC-Boss/GPT-SoVITS README
• GPT-SoVITS v2Pro / v2ProPlus pretrained model notes