Speechify offers a small set of text-to-speech models. Picking the right one is a trade between latency, language coverage, and voice quality. This guide maps each model to the work it fits, so you can choose without testing everything.
The speechify.ai model posts go deeper on each release. The Simba 3.2 announcement covers why it is now the recommended model.
For more on getting started with the Speechify text-to-speech API, check out our quickstart guide.
What models does Speechify offer?
Three families.
- Simba 3.2: the recommended model for English. Streaming-native, lowest time to first byte, curated English voices. Use it for anything interactive.
- Simba 3.0: the current API default. Streaming-native, multilingual: English plus German, Spanish, French, Italian, and Brazilian Portuguese. Slightly higher latency than 3.2 but broader reach.
- Legacy models (simba-english, simba-multilingual): the Simba 1.6 pair. Retired from API version 2026-09-21 and switched off on 2026-11-21. Use them only while an existing integration depends on a specific older voice or language, and plan the migration now.
Which model is fastest?
Simba 3.2. It is built for streaming, so the first audio arrives quickest. For English voice agents, it is the right default.
Simba 3.0 is close but carries multilingual overhead. The legacy pair is the slowest and should be retired where possible.
Which model covers the most languages?
Simba 3.0. It officially supports English plus German, Spanish (Spain and Mexico), French, Italian, and Brazilian Portuguese. Pass the language parameter on POST /v1/audio/speech to select a locale, and it returns a native voice for that language. The legacy simba-multilingual covers the full 30+ locale set, but it is retired from API version 2026-09-21 and switched off on 2026-11-21.
If your product ships in one language, Simba 3.2 wins on speed and quality. If it ships in several, Simba 3.0 wins on coverage right now.
Read more on our language support on our docs.
Which model sounds best?
Quality tracks the model generation. Simba 3.2 leads on English naturalness. Simba 3.0 holds its own across its languages. Legacy models are the oldest and most robotic.
For customer-facing voice, prefer Simba 3.2 or Simba 3.0.
How do I list available voices?
Call GET /v1/voices. Filter by type, locale, gender, and model to find the right voice before you synthesize. The voice and model posts on speechify.ai show the response shape.
Streaming or batch?
Both Simba 3 models stream. Use POST /v1/audio/stream for live playback, POST /v1/audio/stream/with-timestamps for live audio with timestamps and wordmark alignment, and POST /v1/audio/speech for a single file. Model choice is independent of delivery mode.
FAQ
What is the recommended Speechify TTS model?
Simba 3.2. It is the default for new work: streaming-native, fastest to first audio, and the highest quality for English.
Simba 3.2 for English: streaming-native, fastest to first audio, and the highest quality for English. The API default when a request omits model is Simba 3.0, so set model: "simba-3.2" explicitly to opt in.
Yes. Simba 3.0 takes a language parameter and returns native voices across many locales. Simba 3.2 focuses on curated English.
Yes. Simba 3.0 takes a language parameter and returns native voices across English and six European locales. Simba 3.2focuses on curated English.
Only if an existing integration depends on a specific older voice. Otherwise move to Simba 3.2 or Simba 3.0.
Only while an existing integration depends on a specific older voice. They retire from API version 2026-09-21 and switch off on 2026-11-21, so migrate to Simba 3.2 or Simba 3.0 before then.
Pick Simba 3.2 for the lowest time to first byte. Stream the output for live use.
