Voice Library
Every registered voice with its reference transcript. Play the reference clip, edit the transcript, replace the audio, or send a voice straight to the synthesizer.
Analytics
Every generation, metered: audio produced, time-to-first-byte, queue and GPU time, by voice and by API key. Requires the admin key.
Audio generated
By voice
By API key
Time to first byte
Generations
| Time | Key | Voice | Via | Text | Audio | TTFB | RTF |
|---|
Synthesize
One request, one utterance. Great for checking a voice or producing a clip. ⌘/Ctrl + Enter to run.
Input POST /v1/tts
Advanced
Result
Stream
The WebSocket protocol used by LiveKit agents: config → text → flush. Audio arrives per sentence; cancel is barge-in.
Session WS /tts/ws
Collected audio no chunks yet
Protocol log
Clone a voice
5–10 seconds of clean speech and the exact words spoken. The transcript matters: a wrong or partial one breaks pacing; leave it empty rather than guess.
Reference clip POST /v1/voices/upload
Tips for a good clone
- One speaker, no music or background noise, natural pace.
- The server rejects transcripts that can't match the clip (≈ 4–40 characters per second).
- Voices are pushed to the S3 mirror automatically and preloaded on every server start.
- Use the same language in the transcript as in the clip.
API keys & usage
Requires the admin key in the connection settings. Keys authenticate clients; usage is metered per key.
| Name | Key | ID | Status | Expires | |
|---|---|---|---|---|---|
| — | |||||
Server status
Live view of the engine. Refreshes every 5 seconds.