Beta
Arabic text to speech that speaks the right dialect
Note: The model is still being trained, so output quality may vary. For best results, use longer utterances.
Choose a regional voice
Browse the voice library by dialect, listen to samples and pass your chosen voice ID to the speech API.
Start playback while audio is generated
The speech endpoint streams raw PCM, so voice agents and interactive products can play the first audio before the complete response is ready.
Use the SDK you already know
Point an OpenAI-compatible client at the Sawtak Arabi base URL, pass a voice id, and keep the same speech request shape.
From Arabic text to production audio
Use the browser demo to compare voices, then move the same choice into the API. Pricing is per input character, failed server-side requests are refunded, and the response reports the sample rate it actually uses.
- 01
Pick a regional voice
Browse voices by region and dialect, then copy the selected voice id into your request.
- 02
Send Arabic text
Call POST /v1/audio/speech with your text, voice id, and pcm response format.
- 03
Stream and play
Read the PCM response as it arrives and use the X-Sample-Rate header for correct playback.
Questions developers ask
Does the API support Arabic dialects or only Modern Standard Arabic?
It supports both. The shipped library includes voices labelled by region across Gulf, Egyptian, Levantine, Iraqi, Sudanese, Yemeni, and Maghrebi Arabic, plus Modern Standard Arabic.
Can I use the Arabic TTS endpoint for a realtime voice agent?
Yes. The endpoint streams raw 16-bit mono PCM and follows an OpenAI-compatible request shape, which fits frameworks that accept a streaming speech endpoint.
How is Arabic text to speech priced?
Text to speech costs $50 per one million input characters, including spaces, punctuation and diacritics. Accounts use a prepaid USD balance with no monthly subscription.