Every building block for voice, one API.
The developer API for voice AI: text to speech and voice cloning on the same speech stack that powers Speechify, exposed as building blocks you can compose.
50K characters free every month. Commercial use included, no card.
- Text to speech
- Voice cloning
- Sub-300ms first byte
- One key, one bill
curl -X POST https://api.speechify.ai/v1/audio/speech \
-H "Authorization: Bearer $SPEECHIFY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"input": "Ship production voice in a few lines of code.",
"model": "simba-3.2",
"voice_id": "george",
"audio_format": "mp3"
}'Hear it before you write a line.
Type a script, pick a voice, and stream it from Simba 3.2 in the browser. Same request you'll make in production.
First request
Choose parameters inline, then copy the request.
curl -X POST https://api.speechify.ai/v1/audio/speech \
-H "Authorization: Bearer $SPEECHIFY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"input": "The quick brown fox jumped over the lazy dog",
"model": "simba-3.2",
"voice_id": "geffen_32",
"audio_format": "mp3"
}'POST /v1/audio/speech
Run the request, then play the result.
Response appears here after the request completes.
Selected sample: Geffen, neutral.
Everything you need to ship voice.
Text to speech and voice cloning, on one API, one key, and one bill. Start on the free tier and compose them into whatever you're building.
Text to Speech
Generate lifelike speech from text, with streaming, SSML, and emotion on the Simba models.
Voice Cloning
Create a voice from a short sample, get a voice ID, and synthesize with it on the same endpoints.
The #1 model on Artificial Analysis.
The speech block runs on Simba 3.2, #1 on the independent Artificial Analysis TTS leaderboard and above every ElevenLabs, OpenAI, and Google model, at roughly a third of the price of the model closest to it. The result is independent, not self-reported.
Three Simba models, one endpoint.
The models behind text to speech and voice cloning. Pick one with a single parameter; start on Simba 3.2 for new English builds.
Simba 3.2
RecommendedStreaming-native flagship with the lowest time-to-first-byte
Built for real-time playback. First byte in under 300ms, finer-grained emotion control, and SSML prosody, served from a curated voice set. English only, and the recommendation for any English build.
- Streaming-native
- Emotion control
- SSML prosody
- Lowest TTFB
Simba 3.0
Streaming across English and six European languages. Set the language on simba-3.0 and one model covers them all, no model switching, with zero-shot voice cloning.
Simba 1.6
The broad multilingual model (simba-multilingual): 30+ languages and mixed-language input, with zero-shot voice cloning. Non-streaming.
Everything the speech block includes.
The things other vendors charge extra for are already part of the endpoint.
Streaming for realtime
Chunked audio that starts playing before the sentence finishes generating. Up to 20,000 characters a request.
Voice cloning
Create a voice from a short sample, get a voice ID, and synthesize with it on the same endpoints. Consent required.
SSML control
Full SSML (prosody, break, emphasis, sub) for exact control over delivery and timing.
13 emotion styles
Shift a single voice across narration, dialogue, and emotional registers with speechify:style.
Word-level speech marks
Timestamps for every word, returned alongside the audio, for synced captions and read-along highlighting.
Five audio formats
wav, mp3, ogg, aac, and pcm, plus u-law from the stream endpoint for telephony.
One bill for every block.
Text to speech is flat per character, no credits to convert, no token math. Voice cloning bills to the same plan and the same balance.
Built for teams at scale.
SOC 2 Type II, SSO, a signed DPA, and volume pricing, plus a Forward Deployed Engineer who builds alongside heavy-traffic and regulated teams. Most ship a first use case in two to three weeks.
Common questions.
What is SpeechifyAI Build?
The developer API for voice AI, organized as building blocks. Text to speech and voice cloning are available today: generate lifelike speech, stream long-form audio, and clone a voice from a short sample, all on one API, one key, and one bill.
What can I build with it today?
Anything that needs synthesized speech: narration, agents, accessibility, captions, IVR, dubbing. Text to speech and voice cloning are live now on one key and one bill.
Which model should I use?
Simba 3.2 for English, streaming-native with the lowest time-to-first-byte. Simba 3.0 for English plus six European languages: set the language and one model id covers them all. Simba 1.6 (simba-multilingual) for the full 30+ language set. Voice cloning is available on Simba 3.0 and Simba 1.6.
How much does it cost?
Text to speech is from $6 per 1M characters on Scale, $8 on Pro, $10 on Starter, billed per character with no credits or token math. The Free plan includes 50K characters a month with commercial use.
Start building with voice.
Your first audio file is one request away. 50K characters free every month, commercial use included.
No credit card required.