Voice Characters and Voice Banks

Voice characters are Neiroha's main project-level voice abstraction. A provider only describes how to reach a service. A voice character describes the speaker, model, voice, speed, cloning reference, and style used by workflows.

First Setup

1. Finish provider setup and enable at least one provider.
2. Open Voice Bank.
3. Select Default Bank, or click + to create a new bank.
4. Click New Character inside the selected bank.
5. Select the provider, model, voice, or clone mode.
6. Save the character.
7. Type a short sentence in Quick Test on the right and generate audio.

What a Voice Bank Does

A voice bank is a set of voice characters. Projects bind to one voice bank, and every character selector in that project is populated from that bank.

Use Case	Recommended Layout
Single narrator	Put one narrator character in one bank.
Audio drama or game dialogue	Put the full cast in one bank.
Multilingual testing	Split by language, such as Chinese Narration and English Cast.
API server	Put common characters in the active bank so /v1/audio/voices returns them.

Create a Voice Bank

1. Click + New Bank in the Voice Bank page.
2. Enter a bank name.
3. If this bank should be the default API bank, mark it as active.
4. Select this bank when creating Dialogue TTS, Phase TTS, Novel Reader, or Video Dubbing projects.

The active voice bank affects the local API server:

• /v1/models returns active voice banks.
• /v1/audio/voices returns characters in the active voice bank.
• POST /v1/audio/speech can use model to scope voice lookup to one bank.

Create a Voice Character

Open Voice Bank, select a bank, then click New Character. Common fields:

Field	Description
Name	Display and matching name used by the UI, projects, and API requests.
Provider	An enabled provider.
Task mode	Controls whether the character editor shows preset voices, reference audio, or voice instruction fields.
Speed	Synthesis speed multiplier, usually between 0.5 and 2.0.
Avatar	Optional image displayed in dialogue bubbles.

Choose Task Mode by Backend

Mode	Typical Backends	Main Fields
Preset voice	OpenAI-compatible services, Azure, Windows SAPI, Gemini	Select a voice from the provider voice list.
Voice clone with prompt	GPT-SoVITS, CosyVoice, VoxCPM2	Provide reference audio and matching transcript when required.
Voice design	Models that accept voice_instruction or Chat Completions audio output	Describe the desired voice in natural language.

Provider-Specific Character Fields

Provider	Most Important Fields
OpenAI-compatible	Model and preset voice.
MiMo	Model, preset voice, voice instruction, or clone reference depending on mode.
Gemini	Model, Gemini preset voice, optional voice instruction.
Azure	Azure voice short name, such as zh-CN-XiaoxiaoNeural.
GPT-SoVITS	Registered voice, or reference audio plus prompt text.
CosyVoice	Profile, prompt audio, prompt text, or instruction text depending on mode.
VoxCPM2	Registered voice, design instruction, clone audio, or prompt audio.
Windows SAPI	Installed Windows SAPI voice.

If model or voice dropdowns are empty, return to the provider page and click Fetch All. If the backend does not expose list APIs, enter the model or voice name manually.

Test Immediately After Creation

Do not start a long batch right after creating a character. Test one sentence first:

1. Select the new character.
2. Enter one short sentence in Quick Test.
3. Click generate.
4. If audio plays, the provider, character, and queue are usable.
5. Then move to Dialogue TTS, Phase TTS, or Novel Reader.

Troubleshooting

Symptom	Action
Character cannot be saved	Check required fields, especially provider, task mode, voice, and reference audio.
Provider dropdown is empty	The provider is disabled or unsupported on the current platform.
Voice dropdown is empty	Click Fetch All on the provider page, or fill the voice manually.
Quick Test returns 401 / 403	Check API key, Azure region, or cloud account permissions.
Quick Test returns 429	Lower provider concurrency and RPM / RPD limits.
Local clone cannot find reference audio	Make sure the file is reachable from the current device. Android cannot read a Windows local path.