Skip to main content

Quick Start

Requirements

  • Windows 10/11, Linux x64, or an Android device.
  • At least one reachable TTS backend: local, LAN-hosted, or cloud-hosted.
  • Flutter is not required for normal use. Download a Release build instead.

1. Download a Release Build

Open Neiroha Releases and download the latest version. The current latest version is v0.3.1, released on 2026-05-30.

PlatformFileUsage
Windowsneiroha-v0.3.1-windows-x64-release.zipExtract it and run neiroha.exe.
Androidneiroha-v0.3.1-android-release.apkCopy it to the device and install the APK.
Linux x64neiroha-v0.3.1-linux-x64-release.tar.gzExtract it and run the executable in the bundle.

SHA256 checksums are now written directly in the Release page under Checksums. Separate SHA256SUMS*.txt files are no longer provided.

See Install Release Builds for detailed installation steps.

2. Prepare a TTS Backend

Neiroha is a TTS workstation and middleware layer. It does not include large-model inference by itself. Prepare one backend before first use:

ChoiceGood ForNext Step
Local inference backendLocal GPU users, LAN inference servers, or workflows that keep text localConnect Local Inference Backends
Cloud / free quotaQuick trials without local model deploymentConnect Cloud Inference Backends
Windows system voiceWorkflow validation without AI voicesUse Windows System TTS in Providers

Local backend Windows NVIDIA portable packages are available, mainly for RTX 30 / 40 / 50 series users. GPT-SoVITS, VoxCPM2, and CosyVoice3 can be downloaded as split archives from their own Release pages. If GitHub downloads are unstable, use the Baidu Netdisk mirrors in the Release body; see Windows Portable Backend Packages.

Choose a Route

GoalRecommended Route
Hear the first sample as fast as possibleUse Windows System TTS or a cloud free quota, then pass Quick Test.
Keep text localUse a GPT-SoVITS, CosyVoice3, or VoxCPM2 local backend.
Test Chinese-English or multilingual outputTry Gemini, MiMo, CosyVoice3, or VoxCPM2 first, then keep the best provider.
Use reference-audio cloningUse GPT-SoVITS, CosyVoice3, or VoxCPM2 and prepare clean short reference audio.
Batch novels, audiobooks, or subtitle dubbingPrefer a local backend; cloud providers should use RPM, TPM, RPD, and low concurrency limits.
Serve scripts, games, or external toolsCreate a voice bank, pass Quick Test, then enable the API Server.

3. Configure a Provider

Open Providers. The left side lists providers, and the right side shows the selected provider form.

Provider configuration page

Basic flow:

  1. Click + at the top of the provider list.
  2. Select an adapter type.
  3. Fill Base URL, API Key, and any required model name.
  4. Click Fetch All to fetch models and voices.
  5. Enable the provider.
  6. Click Health Check.

For details, see Configure Providers.

4. Create a Voice Bank and Character

Open Voice Bank. Voice banks group characters, and later workflows select voices from the chosen bank.

Voice Bank page

For first use, select Default Bank, then select Default Voice and check the provider, model, and voice binding on the right.

5. Run the First Quick Test

After selecting a character in Voice Bank, the Quick Test panel appears on the right.

Quick TTS page
  1. Type one short test sentence.
  2. Click the purple generate button.
  3. If the provider is configured correctly, audio enters the shared TTS queue and plays automatically.
  4. Generated audio is stored in the voice asset directory and can later be managed by storage scans.

6. Next Steps