Speech synthesis models

Our models are built for different use cases — from expressive English narration to native-quality multilingual synthesis.

Model Recommended

Simba 3.2

model="simba-3.2"

Streaming-native speech with ultra-low latency and rich expressivity

Our flagship streaming-native English model, with the lowest time to first byte in the Simba family, finer-grained emotional control, SSML prosody, and a curated voice set.

<300ms
Latency
English
Languages
Growing
Voices
24kHz
Sample rate
  • Streaming-native architecture
  • Emotional expression
  • Curated voice set
  • SSML prosody control
Demo

Emotion Control

Direct the same line toward neutral, calm, cheerful, energetic, or sad delivery. Simba 3.2 shapes rhythm and tone with SSML emotion control.

Neutral
Calm
Cheerful
Energetic
Sad

“Every moment of light and dark is a miracle.”

Model

Simba 3.0

model="simba-3.0"

Streaming-native synthesis beyond English

Streaming-native synthesis in English, German, Spanish, French, Italian, and Brazilian Portuguese. Routing happens automatically by language, and zero-shot voice cloning works self-serve.

Streaming-native
Synthesis
7
Locales
Self-serve
Voice cloning
24kHz
Sample rate
  • Streaming-native architecture
  • 7 supported locales
  • Zero-shot voice cloning
  • Emotional expression
  • SSML prosody control
Language demo

Multilingual Synthesis

Use Simba 3.0 for English, German, Mexican Spanish, French, Italian, and Brazilian Portuguese. Each sample uses a voice cataloged for that locale.

English

en-US

German

de-DE

Spanish

es-MX

French

fr-FR

Italian

it-IT

Portuguese

pt-BR

Voice demo

Zero-Shot Voice Cloning

Create a reusable voice from a short, consented reference clip. Self-serve cloning on Simba 3.0 preserves the speaker's identity across new scripts.

Reference Original speaker
Clone Simba 3.0 output
Model Legacy

Simba 1.6

model="simba-multilingual"

Legacy multilingual coverage for existing integrations

Our legacy non-streaming model for 30+ locales, mixed-language input, and zero-shot voice cloning. It remains available for existing integrations. For new builds, use Simba 3.0 wherever its validated locale set fits.

<750ms
Latency
30+
Languages
1,500+
Voices
24kHz
Sample rate
  • 30+ languages
  • Mixed-language input
  • Zero-shot voice cloning
  • Existing integrations

All models, one API

Access every model through the same endpoint. Switch between models with a single parameter change.

python
from speechify import Speechify

client = Speechify()  # uses SPEECHIFY_API_KEY env var

response = client.tts.audio.speech(
    input='<speak><speechify:style emotion="cheerful">Every moment of light and dark is a miracle.</speechify:style></speak>',
    voice_id="geffen_32",
    model="simba-3.2",
    audio_format="mp3",
)

with open("output.mp3", "wb") as f:
    f.write(response.audio_data)