Arabic Text to Speech

We provide native voice AI APIs powered by SILMA’s state-of-the-art Arabic TTS models. As Arabic specialists, our models are trained to deliver natural, lifelike AI audio for Modern Standard Arabic (Fus’ha) and the Saudi Najdi dialect - using 8 brand-new native voices.

Convert Arabic text to speech with our Arabic Text-to-Speech (TTS) Model & API. Enjoy the first 5 minutes of audio generation for free.

Generate Arabic AI Speech Now

Built for Voice Agents

Our Text-to-Speech (TTS) models are made specifically for voice-agent companies and designed to work reliably in real production environments.

Our main model, SILMA TTS v2.0, reaches 170 ms TTFT on the server and is trained only on conversational speech. It’s built for large-scale deployment, giving you strong control over speaking speed and how the voice varies, and it also supports voice cloning.

Our Text-to-Speech (TTS) models are made specifically for voice-agent companies and designed to work reliably in real production environments.

Our Voice AI Platform

Text to Speech Generation

Generate speech from Arabic text in Modern Standard Arabic (MSA/Fusha) or the Saudi dialect. Control the speed and speaking style, and choose from a wide selection of different voices.

API Integration

Effortlessly integrate with the platform via robust APIs.

Try our AI Voice Platform Now

SILMA TTS v2.0

~170 ms TTFT

Low streaming latency with ~170 ms TTFT - excluding network

Rock-solid Performance

Superior stability, audio quality and naturalness

On-prem deployment

Flexible hosting options for sensitive customers

Lower Cost

Scale your voice agent platform cost-effectively with our affordable paid plans - starting from $0.025 per minute

Multilingual

We offer native English and Arabic models, with support for code-switching between Modern Standard Arabic (Fusha) and the Saudi Najdi dialect—all within a single, unified pipeline.

Control

Ability to control style variance and speed

Voice cloning

Our models support high quality voice cloning

Voices

10 brand-new voices

Flexible API

We offer multiple streaming options across our TTS APIs, including WebSockets, Server-Sent Events (SSE), and HTTP-based REST.

Lower Cost

Scale your voice agent platform cost-effectively with our affordable paid plans - starting from $0.025 per minute

Multilingual

We offer native English and Arabic models, with support for code-switching between Modern Standard Arabic (Fusha) and the Saudi Najdi dialect—all within a single, unified pipeline.

Control

Ability to control style variance and speed

Voice cloning

Our models support high quality voice cloning

Voices

10 brand-new voices

Flexible API

We offer multiple streaming options across our TTS APIs, including WebSockets, Server-Sent Events (SSE), and HTTP-based REST.

Latency Performance

Below is a comparison of SILMA TTS v2 against leading competitors in terms of latency and pricing, based on recent 2026 benchmark data (with calls placed from the EU after warm-up)

Based on recent 2026 benchmark data (with calls made from the EU after warm-up), here is how the top players stack up in terms of latency and pricing:

Provider / Model	Min TTFT	Median TTFT	Mean TTFT	Cost per Minute
Cartesia Sonic 3.5 (sonic-3.5)	102ms	120ms	123ms	$0.028
ElevenLabs Flash v2.5 (fastest)	131ms	137ms	137ms	$0.050
Deepgram Aura-2 (latest)	251ms	256ms	260ms	$0.027
SILMA TTS v2 (silma-tts-v2-ksa)	279ms	280ms	286ms	$0.025
ElevenLabs v3	677ms	730ms	725ms	$0.100

Try our Voice AI Platform

Integrate our Text to Speech API

Integrating our API is simple: log in to our application, generate your API key, and then follow our API tutorial and specifications.

Integrate Now

API Specifications

In addition, we provide plugins for the following frameworks

Pipecat

pip install pipecat-silma

FAQ

What is the end-to-end TTFT latency, including network time?

Which languages are supported?

Do you support pronouncing phone numbers and email addresses?

What is the maximum input length (number of characters allowed)?

Why SILMA is the best TTS option?

If you’re building a voice agent or integrating an application, we genuinely believe SILMA is the best choice - offering the ideal combination of speed, price, and quality

SILMA TTS = BEST SPEED x BEST QUALITY x BEST PRICE

Try it Now

Why SILMA is the best TTS option?

If you’re building a voice agent or integrating an application, we genuinely believe SILMA is the best choice - offering the ideal combination of speed, price, and quality

SILMA TTS = BEST SPEED x BEST QUALITY x BEST PRICE

Try it Now

Why SILMA is the best TTS option?

If you’re building a voice agent or integrating an application, we genuinely believe SILMA is the best choice - offering the ideal combination of speed, price, and quality

SILMA TTS = BEST SPEED x BEST QUALITY x BEST PRICE

Try it Now

Our open-source TTS Models

SILMA TTS v1 is a powerful bilingual (Arabic/English) text-to-speech model developed by SILMA AI. Built on a cutting-edge F5-TTS diffusion architecture, it delivers high-performance speech synthesis and was trained from scratch using large volumes of high-quality public and proprietary data. To help expand access for the broader community, SILMA TTS is released under a highly permissive license, enabling researchers and businesses to use state-of-the-art TTS technology.

Explore on Hugging Face

Arabic TTS Benchmark

Introducing the Arabic Text to Speech (TTS) Benchmark - a tool for side-by-side comparison of Arabic speech synthesis models

SILMA.AI introduces this benchmark as a primary step toward establishing a gold standard for Arabic TTS evaluation. Recognizing that conventional quantitative metrics (e.g., WER, CER, SIM, UTMOS) often fail to capture the intricacies of the Arabic language, this initial release emphasizes direct auditory assessment. This allows for a genuine evaluation of model naturalness, with more technologically advanced iterations planned as our capabilities evolve.

Explore the Benchmark

Try our Models