Voice Studio creates reusable character voice WAVs and organizes them in a shared local voice library. It can generate with ElevenLabs — Cloud or with a separate, already-running VoiceStudio — Local installation. Saved WAVs can then be previewed, audited, and assigned to a Persona or other voice slot.
Setup Versus Activation
Open Addons → Voice Studio to reach this workspace.
- Addon setup: enable Voice Studio in the Addons workspace and restart NC if requested.
- Provider setup: in Settings, connect either ElevenLabs or your own VoiceStudio installation, then select Save settings.
- Creation: in Create Voices, generate a candidate and explicitly Save selected WAV.
- Activation: assign that saved WAV to the intended Persona or buddy and use a compatible TTS backend.
The addon can be loaded while neither provider is ready. That is a setup problem, not a saved voice.
ElevenLabs Cloud
Open Settings → ElevenLabs — Cloud:
- Use NC ElevenLabs connection uses the existing NC ElevenLabs configuration.
- Session-only API key lets you enter a key for this session without saving it. Prefer the NC connection for normal use.
- Voice Design model chooses
eleven_ttv_v3oreleven_multilingual_ttv_v2for character generation. - Longer-dialogue TTS model is used for the longer-dialogue candidate path.
It is a free-text model ID; the default is
eleven_multilingual_v2. - Check ElevenLabs connection reports readiness.
Cloud generation uses ElevenLabs account credits. The UI explicitly notes that a cost/quota estimate is unavailable, so check your provider account before a large batch. Preview WAVs do not occupy saved voice slots; the saved library is separate.
If a request outcome is unknown, do not blindly generate again. Review your ElevenLabs account and use the explicit retry or resolution control. NC avoids guessing a voice ID or repeating a paid request automatically.
VoiceStudio Local
The local option connects to a separate VoiceStudio installation:
- Install, start, and load an engine in that application yourself.
- In NC Settings, confirm the local URL, normally
http://127.0.0.1:3900. - Select Check local connection.
- Review the detected engines and whether each is loaded.
- Select Save settings.
NC does not install packages, download models, start the service, or load its engines. If the status says an engine must be loaded manually, do that in the VoiceStudio application.
The supported design engines are OmniVoice and VoxCPM2 when the local service reports them. OmniVoice uses a fixed tag vocabulary, while VoxCPM2 takes freeform instructions.
OmniVoice accepts only a limited set of gender, age, pitch, whisper, and English accent tags. Choose one attribute per category; other traits remain in the character description. English accent tags require English speech. Unsupported tags are rejected rather than silently dropped. Each local candidate is a separate generation.
Create A Voice
In Create Voices, choose the Creation provider and either add characters manually or use Draft characters with AI.
To draft a cast:
- Describe What voices do you need?
- Set Voice count.
- Optionally leave Automatic — save first valid candidate off so you can compare candidates.
- Select Draft characters with AI.
- Review the generated names, descriptions, voice descriptions, and sample text.
- Select a character, then Generate selected.
AI drafting and refinement use the active NC LLM. They do not generate the final voice.
Preview and save
When candidates exist:
- Preview auditions the selected candidate.
- Stop ends the preview or batch.
- Save selected WAV stores the validated candidate in the local library.
- Regenerate / retry creates another candidate from the same description.
- Allow retry after checking provider appears for an unknown outcome after you have checked the provider account.
A candidate is not a library voice until it is saved. Saving a new file does not overwrite an existing one, and filenames are made unique.
Save cast drafts stores the editable cast descriptions and prompts. It does not save audio. Load saved drafts restores that editable work, not generated WAVs. Remove selected draft removes the draft character; existing recordings stay in the library.
Longer ElevenLabs Dialogue
The Longer dialogue — ElevenLabs candidate section can record new text using a tracked temporary cloud voice. Generate dialogue and clean up verifies the local WAV before deleting that temporary voice. Credits and account operations still apply.
If cleanup is pending, Temporary cloud voices in Settings offers Retry cleanup and, where applicable, an explicit option to discard unsaved dialogue and delete the tracked voice. Do not delete the temporary voice manually while an NC save is pending.
Voice Library
Voice Library lists saved local recordings, including WAVs created elsewhere and shared with Persona.
- Select a voice to inspect its path, duration, source, and metadata.
- Play original sample plays the file.
- Description and Notes are shared with Persona and do not alter the recording. Select Save details.
- Remove from library hides the entry from both libraries but leaves the WAV in place so saved Personas keep working. Show removed voices and Restore to library make it visible again.
- Open voices folder opens the managed library folder.
Removing a library entry is not the same as deleting the file. Voice Studio keeps the original so existing Persona assignments do not break silently.
Audition Your Own Words
In Voice Library:
- Load Chatterbox or Chatterbox Multilingual in Runtime → Runtime Session → Text-to-speech.
- Select a valid saved voice.
- Enter text under Try your own words.
- Select Generate test, then Play test.
- Optionally select Save test recording… to store a separate WAV. This export creates a new file and refuses to overwrite an existing path.
Audition tests preserve the normal chat voice and stay separate from reference recordings. Original samples can still be played without the Chatterbox audition runtime. Refine with AI changes test text, not the voice file.
Assign The Finished Reference
Saving a WAV in Voice Studio does not automatically change the main chat voice or attach it to a Persona. To use it:
- Open the intended Persona’s Voice tab, or the relevant Buddy Chat voice setting.
- Select the saved WAV from the shared library.
- Preview the original if needed.
- Select the control that saves the selection to that Persona.
- Use a TTS backend that supports reference audio.
- Test one short character reply.
The Persona library can show Assigned for a file, but the assignment still belongs to the Persona’s saved voice settings. Saving the library entry alone is not enough.
Batch Behavior
- Pause batch stops the current batch after the active operation.
- Stop cancels the current work; it does not roll back a candidate that was already saved.
- Automatic — save first valid candidate bypasses manual candidate selection. Leave it off when you want to compare takes.
- Generate all / continue processes the cast. Review provider cost before a large batch.
- A provider or local service error can leave a character in a review or unknown state. Inspect that state before retrying.
Troubleshooting
| Symptom | Check |
|---|---|
| ElevenLabs is not ready | Select Check ElevenLabs connection, confirm the NC connection or session key, and check account access |
| Local service is not connected | Start the separate VoiceStudio installation, load its engine, confirm the URL, and select Check local connection |
| A voice will not generate | Check the provider, character description, sample text, and any per-engine tag limits |
| A test cannot be played later | Generate test and Play test are session auditions; use Save test recording… to keep the result |
| The voice did not change in chat | Assign the WAV to the Persona or buddy, use a reference-capable TTS backend, and test a short reply |
| A saved entry disappeared | Enable Show removed voices; removal hides the entry but keeps the WAV |
| A cloud request outcome is unknown | Check the ElevenLabs account and use the explicit resolution/retry path; do not repeat a paid request blindly |
| Temporary voice cleanup is pending | Use Retry cleanup in Settings and keep the tracked record until local saving is verified |
Current Limits
- NC does not install, launch, or manage the separate VoiceStudio local service.
- Cloud generation needs valid ElevenLabs credentials and available credits. NC cannot estimate the exact cost or quota.
- Voice quality depends on the provider, description, text, and model. A saved WAV is a reference, not a guarantee that a TTS backend will reproduce it perfectly.
- The Voice Studio library and Persona assignment are separate. A saved WAV is not automatically the main chat voice.
- Removing a library entry does not delete the underlying file.
- Batch generation and longer dialogue can consume significant provider quota.
- OmniVoice accepts only its fixed tag vocabulary; unsupported traits must stay in the voice description rather than the instruction.
