The science of speech. The character of a voice.

Build on Simba 3.5, our flagship: 48 languages under one model id, with every catalog voice in each. The older Simba 3.2 (English) and Simba 3.0 (6 languages) remain available on the same endpoints.

Hear the Simba 3.5 story in all 48 languages Read the launch post and how to switch

Shared foundation

Streaming-native
Self-serve voice cloning
24kHz output

Speech synthesis model Recommended

Simba 3.5

model="simba-3.5"

48 languages under one model id

Our recommended model since 21 October 2026. One model id speaks 48 languages on every plan, through the same speech, streaming and WebSocket endpoints, with every voice in the catalog and your cloned voices.

Read the Simba 3.5 announcement

Delivery: Streaming-native
Languages: 48
Voice cloning: Self-serve
Sample rate: 24kHz

Capabilities

48 languages, chosen per request
Every voice in the catalog, cloned voices included
Speech, streaming, timestamps and WebSocket
SSML pauses and speaking rate

Measured

Preferred in blind listening

The Simba 3.5 fact sheet's English results against Simba 3.2, as of October 2026. Fact sheet (PDF)

63%: win rate against Simba 3.2 in blind side-by-side listening, all content (95% CI 60-67%)
1.8×: Simba 3.2’s speech throughput per serving instance
72 ms: time to first audio for 95% of streaming requests, under production load

Language range

48 languages, one model id

Pass model: "simba-3.5" and set language to the language of your text, for example ja-JP. Any regional variant of a listed language is in scope, and every catalog voice can speak each of them.

Simba 3.5 sounds best with a voice native to the language you generate. Every voice keeps the accent of the recording it was made from, so an English voice reading German sounds like an English speaker speaking German. A cloned voice works the same way: clone from a recording in the language you will use it in.

We kept this on purpose. Customers asked that a voice, and above all their own clone, keeps sounding like itself in every language. We plan to make this more flexible in an upcoming release.

For each language, start with its top voices.

Price per 1M characters

Starter: ~~List price $15~~ launch price $10
Pro: ~~List price $12~~ launch price $8
Scale: ~~List price $10~~ launch price $6

Launch price until 21 Nov. Simba 3.0 and 3.2 bill $10, $8 and $6. See every plan

languages on Simba 3.5

simba-3.5

Arabicar
Belarusianbe
Bengalibn
Bosnianbs
Burmesemy
Catalanca
Chinesezh
Croatianhr
Czechcs
Danishda
Dutchnl
Englishen
Finnishfi
Frenchfr
Georgianka
Germande
Greekel
Gujaratigu
Hebrewhe
Hindihi
Hungarianhu
Indonesianid
Italianit
Japaneseja
Kazakhkk
Koreanko
Latinla
Malayms
Marathimr
Norwegianno
Persianfa
Polishpl
Portuguesept
Romanianro
Russianru
Serbiansr
Slovaksk
Sloveniansl
Spanishes
Swedishsv
Tagalogtl
Tamilta
Telugute
Thaith
Turkishtr
Ukrainianuk
Urduur
Vietnamesevi

Voice study

One line, 6 voices

The same plain-text line on Simba 3.5 from 6 of the catalog's American and British English voices, each passed by a blind native listener. Simba 3.5 serves every voice in the catalog, your cloned voices included.

Same text · 6 voicessimba-3.5

“The quick brown fox jumped over the lazy dog.”

Dominic01

Geffen02

George03

Hugh04

Imogen05

Wyatt06

Voice cloning

Zero-Shot Voice Cloning

Create a reusable voice from a short, consented reference clip, then give it new scripts on Simba 3.5 in any of its 48 languages. A clone keeps the accent of its recording, so record in the language you will generate.

Explore voice cloning

Reference and synthesissimba-3.5 · zero-shot

01 / Reference

Original speaker

Ready to listen0:00

02 / Clone

Simba 3.5 output

Ready to listen0:00

Two different recordings. Only clone a voice you have permission to use.

Speech synthesis model Older model

Simba 3.2

model="simba-3.2"

The English model, streaming-native

An older model that remains available for English. Audio streams as it generates, with SSML pauses and speaking rate, and every English voice in the catalog, your cloned voices included.

Read the Simba 3.2 announcement

Delivery: Streaming-native
Languages: English
Voice cloning: Self-serve
Sample rate: 24kHz

Capabilities

Streaming-native architecture
Every English voice in the catalog
Self-serve voice cloning
SSML pauses and speaking rate

New integrations

Simba 3.5 reads English too, with every voice in the catalog. To move, change model to simba-3.5; the endpoints, voices and request stay the same.

Simba 3.5

Speech synthesis model Older model

Simba 3.0

model="simba-3.0"

6 languages across 7 locales

An older model that remains available, in English, German, Spanish, French, Italian, and Brazilian Portuguese. It is the API default when a request names no model, so a new integration names Simba 3.5.

Delivery: Streaming-native
Locales: 7
Voice cloning: Self-serve
Sample rate: 24kHz

Capabilities

Streaming-native architecture
7 supported locales
Zero-shot voice cloning
SSML pauses and speaking rate

Language range

6 languages, 7 locales

Simba 3.0 routes each request by language across the locales on the right. Check each voice's supported models in GET /v1/voices before synthesis.

Need a language outside these six? Simba 3.5 speaks 48 under one model id: see its language range.

Locale decksimba-3.0

languages · 7 locales

enEnglish
de-DEGerman
es-MXSpanish (Mexico)
es-ESSpanish (Spain)
fr-FRFrench
it-ITItalian
pt-BRPortuguese (BR)

One synthesis surface

All models, one API

Access every model through the same endpoint. Choose the model and a voice compatible with its language support.

simba-3.5: The flagship: 48 languages under one model id, every catalog voice
simba-3.2: Older model: English, every English voice in the catalog
simba-3.0: Older model: 6 languages across 7 locales, and the default when a request names no model

Text-to-Speech API Documentation

python

import base64

from speechify import Speechify

client = Speechify()  # uses SPEECHIFY_API_KEY env var

response = client.audio.speech(
    input='<speak>Every moment <break time="300ms" /> is a miracle.</speak>',
    voice_id="geffen_32",
    model="simba-3.5",
    audio_format="mp3",
)

# audio_data is Base64-encoded audio bytes
with open("output.mp3", "wb") as f:
    f.write(base64.b64decode(response.audio_data))

Built on speech. Open to possibility.

Make something
worth listening to.

Start building Talk to sales

The science of speech. The character of a voice.

Preferred in blind listening

48 languages, one model id

One line, 6 voices

Zero-Shot Voice Cloning

6 languages, 7 locales

All models, one API

Make somethingworth listening to.

Make something
worth listening to.