60 languages, 22 of them Indic. Clone a voice from 30 seconds of audio and stream it in real time from one API call. Try it below, no sign-up.
- languages
- 60
- languages
- Indic languages
- 22
- Indic languages
- of audio to clone
- 30s
- of audio to clone
Live · no sign-up
অসমীয়াAssamese
Quiet, formal courtesy.
English
A real conversational hitch mid-sentence, not a clean read.
हिन्दीHindi
Calm, unhurried. The question lifts at the end without tipping into sing-song.
12 languages here, 60 in the API · up to 300 characters · 6 plays a minute
- हिन्दी — Hindi
- বাংলা — Bengali
- తెలుగు — Telugu
- தமிழ் — Tamil
- मराठी — Marathi
- ಕನ್ನಡ — Kannada
- മലയാളം — Malayalam
- ગુજરાતી — Gujarati
- ਪੰਜਾਬੀ — Punjabi
- ଓଡ଼ିଆ — Odia
- অসমীয়া — Assamese
- नेपाली — Nepali
- संस्कृतम् — Sanskrit
- मैथिली — Maithili
- कोंकणी — Konkani
- डोगरी — Dogri
- बोडो — Bodo
- मणिपुरी — Manipuri
plus Urdu, Kashmiri, Sindhi and Santali — and 36 global languages
one clip in — sixty languages out
60
Languages
36 global, 22 Indic, 2 CJK
22
Indic languages
down to Bodo, Dogri, Santali
24 kHz
Native output
WAV or raw PCM, 5 encodings
10,000
Characters per request
one call, no chunking
22 of 60 — Indic
Streaming over REST, SSE, or WebSocket — up to four concurrent contexts per connection.
Three steps to a voice that’s yours.
- 01
Record 30 seconds
Upload a clip or record in the browser. Phonetically varied passages work best — questions, numbers, a little emphasis. 20 MB cap.
- 02
Clone
We return a job id and clone asynchronously. Poll it until the voice reads ready, then it's yours to synthesize with.
- 03
Stream
One POST for a finished file, or open a WebSocket and feed it text as your LLM produces it. Audio starts before the sentence ends.
One engine, six very different jobs.
Long-form narration
Audiobooks, courseware, and explainers. Send up to 10,000 characters in a single request and get 24 kHz WAV back — no chunking, no stitching seams to hide.
Conversational agents
Support lines and in-app assistants that answer in the caller's own language. One connection carries up to four concurrent contexts, so a single agent can hold several turns at once.
Realtime streaming
Feed text to a WebSocket as your LLM produces it and audio starts before the sentence is finished. The stream is the point — you are not waiting on a file.
Dubbing and localization
One cloned voice, sixty languages. Language is a per-request field, not a property of the voice — so the same speaker reads your Hindi, your Tamil and your Spanish without re-cloning anything.
Telephony and IVR
Headerless PCM in five encodings at whatever rate your switch expects, plus 24 kHz native. Hand the bytes straight to your telephony stack — there's no container to strip and no transcode step to pay for.
One voice everywhere
Clone once from thirty seconds and reference the same voice id from every surface you ship. Cloning and enrollment are free and cost no characters, so a consistent voice is a decision, not a budget line.
Don't take our word for it
Hear Finesse against the others, live.
The same line in multiple engines, side by side, in your browser.
Consent on every clone
Cloning won't start until you confirm you have the rights and consent to use the voice you upload.
Your voices stay in your workspace
Cloned voices and API keys are scoped to the workspace that created them — not shared across accounts.
Delete a voice anytime
Remove a voice you own from its page in the dashboard, whenever you choose.
Start free with 30,000 characters — about 33 min of speech. Plans from $19/month.
See pricing