Inference APIs
Text-to-Speech API/Voices by language

Text-to-speech voices by language

Kokoro 82M has 54 voices across eight languages; Orpheus 3B adds eight expressive English voices. Every voice has a sample you can play without signing in, generated through this API with default settings.

English (American and British voices)
28 Kokoro voices + 8 Orpheus voices
af_heartall voices
Japanese
5 Kokoro voices
jf_alphaall voices
Mandarin Chinese
8 Kokoro voices
zf_xiaobeiall voices
Spanish
3 Kokoro voices
ef_doraall voices
French
1 Kokoro voice
ff_siwisall voices
Hindi
4 Kokoro voices
hf_alphaall voices
Italian
2 Kokoro voices
if_saraall voices
Portuguese (Brazilian voices)
3 Kokoro voices
pf_doraall voices

How the language is chosen

The voice decides the language. Each Kokoro voice id starts with a language letter (a American English, b British English, j Japanese, z Mandarin, e Spanish, f French, h Hindi, i Italian, p Brazilian Portuguese) and a gender letter, f or m. Send text in the voice's language. An optional language field is accepted as a pronunciation hint, but it does not make an English voice speak Japanese.

What to expect from each language

English is Kokoro's strongest language by a distance: most of its training audio was English, and its authors rate those voices highest. The other languages have fewer voices and less training data behind them, and French has a single voice. As a sanity check we transcribed every sample on these pages with Whisper and compared the result with the text we sent; the per-language pages show the outcome for each voice. Languages that are not listed (German, Korean, Arabic, Russian and others) have no voice yet. If you need one, tell us which.

Price and limits

Kokoro costs $5.20 per million characters and Orpheus $19.50, billed per Unicode character of input, so Japanese and Chinese text is not penalised for multi-byte encoding. A request takes up to 20,000 characters and returns mp3, wav or raw PCM. The voice list is also available from GET /v1/voices?model=hexgrad/Kokoro-82M. Full details: Text-to-Speech API.