Help center
How everything works
Text to speech (studio)
Paste text, pick a voice, optionally pick the model, speed and format, then Generate. Signed-in renders are saved to your history; visitors can use the free anonymous tier which is Turnstile-gated and capped.
Multi-speaker dialogue: add speaker slots, tag each line in the text or use the slots; the model reads the tags. Only models marked multi-speaker support this.
Voice cloning
Record or upload 10–30s of clean speech, confirm consent that the voice is yours or authorized, and submit. The model is trained as private — only your account can call it.
Deleting a clone removes the provider model, the uploaded sample in R2, and its voice-asset row in one operation. We block obvious impersonation prompts before the provider sees them.
Voice design
Describe the voice (age, tone, accent, pace, energy). We generate a few short candidate previews; audition them and save the one you like. The saved model is private and marked as designed, not cloned.
Candidates that fail provider generation are discarded — you never get an empty voice in the picker.
Speech to text
Upload WAV/MP3/FLAC or a video container (≤100MB). The transcript comes back as editable, timestamped segments; click a timestamp to seek the player, edit text inline, and export TXT/SRT/VTT/JSON.
The audio file itself is forwarded to the transcription provider and not kept in our storage — only the transcript text and timestamps persist, and deleting removes them completely.
Voice changer
Upload or record your voice, pick a target voice and convert. The converter transcribes then re-synthesizes, so words and timing follow the transcript; tone/emotion are approximated — a real limitation, not a bug.
Your source recording stays client-side and is discarded after the transcript returns.
Audiobooks & long text
Paste a long script; we split it into ordered narration segments. Assign a voice per segment and render each part one at a time — every segment spends from the same credit pool, so a failed paragraph never burns the rest of your budget.
Edit text or voice per segment any time; progress persists in your library so you can come back and resume.
Video dubbing
Upload a video (≤200MB). We keep the file in your storage for playback and deletion, transcribe the speech into an editable line script, and let you re-render each line with a chosen voice.
Speaker separation and lip-sync are not provided — this is a re-dub pipeline: new voice track + subtitles, exported per line and as SRT/VTT.
Developer API
Create keys on the API keys page. Call POST /api/v1/tts with `x-api-key: <key>` and JSON `{ text, voice, model?, speed?, format? }`; the response is `{ id, audio_url, format, model, characters }`. GET /api/v1/tts returns the OpenAPI 3.0 schema for import into API tools.
API usage draws from the same credit pool as the studio — your quota page shows both. Keys are stored hashed and shown once at creation. Session cookies are not accepted on the public API; key only.
Data & privacy
TTS text is sent to the speech provider only to render; render history (audio) can be deleted per item or in bulk. Voice samples are deleted with their clone; transcripts and projects cascade-delete their segments and media.
We do not train our own models on your content. Feature flags gate every provider-backed surface, so anything listed here is actually on.
Report abuse
To report impersonation, a non-consensual voice clone, fraud, or other misuse of generated audio, use the contact page or the support email in the footer. Include the audio or voice name and where you found it.
The site operator reviews every report and can suspend the account and delete the offending voice model and its assets. Expect a first response within two business days.