/voice-cloning and /voice-design routes redirect to /voices.
What You Can Do
Using the Web UI
- Open
http://localhost:8080. - Select Voices in the sidebar.
- Use the Saved, Built-in, or All views to find a voice.
- Select New Voice to create a cloned or designed voice.
- Select Use in TTS to continue generation in Text to Speech.
Saved Voices
Saved voices are reusable voice assets stored by the local server. A saved voice can come from:- A reference-audio voice clone
- A prompt-designed voice
- A generated voice source saved through the API
/v1/voices, and izwi tts can reuse a saved voice with
--saved-voice-id.
A cloned voice can be saved without its transcript. Such an audio-only voice
works with Chatterbox-500M and Chatterbox-Multilingual-500M, which clone from audio alone; models that need a
transcript (Qwen3-TTS Base, VibeVoice, Fish Audio S2 Pro, Fish Audio Drama 3
and Maya 2 Native) refuse it with a 400 asking for one. A saved voice can
also store the language of its recording (reference_language); Maya 2 Native
reads it when the voice speaks any of its 12 languages.
Fish Audio Drama 3 also speaks without any saved voice or reference: its
sampling seed then sets the voice, and the same seed gives the same voice.
Built-In Voices
Built-in voices are model-provided speaker presets. Voice Studio shows the available speakers for the selected model, can generate short preview samples, and can send a speaker into Text to Speech. Use built-in voices when you want quick, repeatable presets without reference audio or a design prompt. Use saved voices when you want a custom reusable voice asset. See Voice Presets for the current speaker IDs.Creating New Voices
Select New Voice from Voice Studio to choose a creation flow:
See the dedicated guides for details:
API Routes
See the API Reference for exact saved voice fields and
the Audio Speech API for generation inputs.