capability
specimens from the production API
One voice.23 languages.
Clone a voice from a ten-second reference and the same person answers in every language on this page — including ones the reference never spoke, verified in blind listening panels. The clips below are production API output, rendered 2026-07-16 and left as they returned.
The same sentence, eight scripts.
01 · the proof you can press
One stock voice — Jenny — reads the same dispatch line through eight writing systems. Press two in a row: the alphabet changes, the person does not.
“Your driver is eight minutes away and dock four is clear — the paperwork is already in your inbox.”
- voice
- Jenny
- language
- English
- script
- Latin
- direction
- left-to-right
- duration
- 5.5 s
The identity is the constant; the language is a request field.
8 of 23 · rendered 2026-07-16 · unedited
Every language rides the same line.
02 · the same budget everywhere
one engine
No per-language model to route, warm, or pay for — the same synthesis path answers in all 23.
one request field
"language": "ar"
A language is a setting on the request, not a plan — nothing changes when a caller switches mid-conversation.
one latency figure
107 / 108 ms
First audio, p50 / p95 single stream, measured on the production API — the datasheet publishes one figure, not one per language.
All 23, set in their own scripts.
03 · the coverage sheet
Eight writing systems, two directions, one identity. Languages with a published specimen play right here; the rest run on the live call — same reference, same person.
English
Latin
specimen · 5.5 s
Español
Spanish · Latin
specimen · 7.1 s
Français
French · Latin
specimen · 5.9 s
Deutsch
German · Latin
specimen · 6.4 s
Português
Portuguese · Latin
specimen · 6.3 s
Italiano
Italian · Latin
on the call
Nederlands
Dutch · Latin
on the call
Polski
Polish · Latin
on the call
Türkçe
Turkish · Latin
on the call
Svenska
Swedish · Latin
on the call
Dansk
Danish · Latin
on the call
Norsk
Norwegian · Latin
on the call
Suomi
Finnish · Latin
on the call
Čeština
Czech · Latin
on the call
Română
Romanian · Latin
on the call
Русский
Russian · Cyrillic
on the call
Українська
Ukrainian · Cyrillic
on the call
Ελληνικά
Greek · Greek
on the call
العربية
Arabic · Arabic · rtl
specimen · 7.6 s
हिन्दी
Hindi · Devanagari
on the call
中文
Chinese · Han
specimen · 6.1 s
日本語
Japanese · Kana · Kanji
specimen · 4.9 s
한국어
Korean · Hangul
on the call
8 of the 23 play dated specimens above — production API output, rendered 2026-07-16, unedited. The other 15 run on the live demo call. All 23 ride one flat line — a language is a setting, not a meter.
Bring ten seconds of your own voice — hear it answer in a language you do not speak.