Voices
Cloning voices for GPT-SoVITS, F5-TTS and Omni/IndexTTS-2 — a clip uploaded here is offered by all of them, whatever the engine label says. Upload a clean voice clip of any length — the app automatically extracts a ~9 second window (from the start time you choose), since the engines need a 3–10s reference. Provide the transcript for that window.
Getting good quality
- Clean reference audio: one speaker, no music/background noise, no reverb. A studio/voiceover clip works far better than a noisy phone recording — noise is the main cause of a "tinny/mechanical" clone.
- Transcript must exactly match the ~9s window (the part starting at your chosen start time), including punctuation. A mismatch makes it sound like a different/"off" person.
- Pick a window where the speaker talks naturally and clearly with normal pacing.
- Words randomly dragging out? That's a GPT-SoVITS timing artifact — the server defaults now reduce it; you can tune further via the
SOVITS_*env vars (see README).
Add a voice
The ~9s reference window starts here. Use it to skip silence/noise at the beginning.
Existing voices
Achernar en · Omni
ready
no transcript
Aoede en · Omni
ready
no transcript
Autonoe en · Omni
ready
no transcript
Callirrhoe en · Omni
ready
no transcript
Charon en · Omni
ready
no transcript
Fenrir en · Omni
ready
no transcript
Kore en · Omni
ready
no transcript
Leda en · Omni
ready
no transcript
Orus en · Omni
ready
no transcript
Puck en · Omni
ready
no transcript
Sadie en · GPT-SoVITS
ready
Hello, and thanks for listening.
Today, I'd like to share a few thoughts about curiosity and the small habits that shape our everyday lives.
sadie-excited en · GPT-SoVITS
ready
Hello, and thanks for listening.
Today, I'd like to share a few thoughts about curiosity and the small habits that shape our everyday lives.
Zephyr en · Omni
ready
no transcript