- Hakim Arab v2 · STT90ms
- Hakim Saree' v1.3 · TTS113ms
- 30-day uptime99.97%
Voice AI that sounds like your world.
The voice AI layerfor the Arabic internet.
Create natural speech, transcribe conversations, and clone voices.
Built for Arabic. Ready for the world.
No card. 30 seconds to first synthesis.
Try it yourself
Type a line. Hear Hakim read it back.
Direct call to the same production pipeline your app will use. 5 free generations per day, no sign-up, no card.
Hear the range
Twelve voices. Fifteen Arabic dialects. One model family.
Tap a voice. We stream a six-second sample from the same endpoint your app will call. No sign-up.
Layla . Modern Standard . Narrative
Khalid . Modern Standard . Announcer
Reem . Khaleeji . Conversational
Yusuf . Egyptian . Conversational
Nour . Levantine . Social
Amir . Maghrebi . Conversational
Omar . Iraqi . Narrative
Salma . Sudanese . Conversational
Amelia . English · US . Conversational
Edward . English · UK . Announcer
Isabelle . French . Narrative
Mateo . Spanish . Conversational
Diego . Spanish · LATAM . Conversational
Kaique . Portuguese · BR . Narrative
Priya . Hindi . Conversational
Felix . German . Conversational
The voices, technology, and trust behind Hakim
People online with limited typing fluency
3.4Bapprox. global
Sources: ITU global internet estimate, Ethnologue first-language speakers, regional typing-literacy surveys. Full references in the launch thesis post.
Voice is the next interface.
The web has grown to five billion people. Typing, in your first language, without a keyboard that fights you, has not. The gap between 'has a phone' and 'can type comfortably' is where voice belongs.
How the Hakim TTS family is built.
Arabic-first TTS in three tiers: Saree' for streaming speed, Faseeh for premium quality with non-verbal tags, Mulhim for voice generation from a free-form prompt. All three tuned by native speakers. The first byte of streaming audio lands in roughly 113 ms p95 on the speed tier.
Hakim Saree' (سريع, swift) is the speed tier of the family · the streaming endpoint tuned for the lowest first-byte latency. It reads Arabic script and diacritics natively, handles English inside an Arabic sentence without switching voices, and exposes the same endpoint whether you pick a region in the EU or the GCC. Hakim Faseeh v2 layers on premium quality plus non-verbal tags ([laugh], [breath], [whisper]) for narration and branded content. Hakim Mulhim v3 adds voice design from a free-form prompt so you can conjure a fresh timbre without recording a single sample. One API, three tiers, same authentication and billing primitives.
113msTime-to-first-audio p95 . Rolling 15-second window across live regions
Every voice is recorded and QA'd with a native speaker before it ships. Evaluation methodology and the full three-tier capability matrix live on the models page.
MSA + 14 regional dialects
Non-Arabic supported range
GCC + EU
Arabic isn't one language. Learn how Hakim Arab v2 is built.
Arabic-first streaming STT. Dialect-aware, code-switch-aware, interactive latency, first token in roughly 90 ms p95.
A Khaleeji voice Hakim Arab v2 is designed to transcribe. Play, then imagine the same accuracy going the other way.
First-token from speech-in
Hakim Arab v2 supported range
Available today
No accuracy or word-error-rate claim is published on this page, we hold that conversation to the models page, where the evaluation protocol is documented.
It's Modern Standard Arabic plus fourteen mutually-semi-intelligible regional dialects, sounds that no English-first model was trained on.
Supported dialects
- MSA
- Khaleeji
- Egyptian
- Levantine
- Maghrebi
- Iraqi
- Sudanese
- Yemeni
- Najdi
- ع/ʕ/voiced pharyngeal
- ح/ħ/voiceless pharyngeal
- خ/x/velar fricative
- ق/q/uvular stop
- ص/sˤ/emphatic s
Supported range refers to current coverage in the Hakim TTS family (Saree' / Faseeh / Mulhim) and Hakim Arab v2; see the models page for the full capability matrix.
How we stack up
Built for Arabic, benchmarked against the world.
We ship voice infrastructure for the Arabic internet, so the comparison that matters runs on Arabic content, not English-tuned demos. Numbers below are calibrated against the latest production tier each vendor offers, measured the week of launch.
| AzureCognitive Services | HakimBuilt for Arabic | ElevenLabsMultilingual v2 + Flash | GoogleCloud Text-to-Speech | OpenAItts-1 + Whisper | Fish Audiofish-speech v1.5 | |
|---|---|---|---|---|---|---|
| Arabic dialects with named voicesCounted from each vendor's TTS catalogue, not generic 'Arabic' fallbacks. | 3 | 15 | 1 | 2 | 1 | 1 |
| Arabic STT word error rateLower is better. 7-hour Arabic-LibriSpeech eval, MSA + four dialects. | Trails | Leads | — | Trails | Competitive | — |
| Voice cloning quality on ArabicHow the vendor handles a 30-second Arabic sample, accent, register, dialect. | Enterprise | Native | English-tuned | Limited | — | No consent gate |
15 Arabic dialects shipped
MSA + 14 regional varieties with named voices
113 ms TTS time-to-first-byte, p50
Arabic content from the region you pick, production checkpoint
leads Word error rate on Arabic STT
Arabic-LibriSpeech eval set, MSA plus four dialects
Hakim leads on 6 of 9 dimensions and ties on 3 more.
Numbers from each vendor's public documentation and our 2026-Q1 internal benchmarks against their highest-quality production tier (ElevenLabs Flash, Azure Neural HD, Google Studio, OpenAI tts-1, Fish Audio real-time). Arabic STT WER measured on a 7-hour Arabic-LibriSpeech evaluation set covering MSA plus four major dialects. Vendor brand names are the trademarks of their respective owners.
FAQ
What people ask before they integrate.
The questions our sales and support teams hear most often.
Built for enterprise
Your audio stays where you chose.
Four live regions. GCC (UAE, KSA, Qatar) plus EU Frankfurt. You pick where your audio is processed; it doesn't leave that region. Your data stays where your customers and regulators expect it to stay. Pick a region when you create a project, and every request, synthesis, transcription, storage, is handled inside that region. No silent cross-border routing, no surprises in your audit trail.
- FrankfurtEU · live
- DohaGCC · live
- DubaiGCC · live
- RiyadhGCC · live
Our connected cities
Our compliance posture today. No badge on this page implies a completed auditor attestation unless the security page's control mapping says so explicitly.
Read the security pageWhat you can build
Three Arabic-first voice models. One API.
Text-to-speech, speech-to-text, and voice cloning, all native-quality across Arabic dialects and 30+ languages, all behind the same authentication and billing primitives.
- See TTS capabilities
Text to speech, Arabic-first.
Arabic-first streaming TTS at 113 ms p95 time-to-first-audio. MSA plus 14 regional Arabic dialects and 30+ other languages out of the box, served from your choice of four live regions.
- See STT capabilities
Speech to text, dialect-aware.
Arabic-first streaming STT. Fifteen Arabic dialects, English and Arabic in the same sentence, 90 ms p95 first token, delivered from your choice of four live regions.
- See voice cloning
Clone any voice instantly.
Record ten seconds, get a production voice that follows your clone's dialect and register. Cloning is instant, and custom voices share the same inference stack as Hakim Fast v1.3, same latency, same quality bar.
Ship in minutes
Five lines to a voice.
Drop-in SDK for Node and Python, plus pure cURL and browser-JS snippets. Every example hits the same pipeline the paid API uses, no separate staging path.
Hakim Fast v1.3 · default TTS model · streaming, low-latency, 15 dialects
Reem · Khaleeji · Khaleeji preset · one of twelve voices shipped with every account
import { Hakim } from '@tryhakim/voice';
const client = new Hakim({ apiKey: process.env.HAKIM_API_KEY });
const audio = await client.audio.speech.create({
model: 'hakim-fast-v1',
voice: 'reem-khaleeji',
input: 'أهلاً وسهلاً بكم في حكيم.',
format: 'mp3',
});
await audio.writeToFile('hello.mp3');Pricing
Pricing that scales with you.
Free while you prototype. Pay only for what you use once you ship. Enterprise terms on request, with data residency, custom retention, and signed BAAs where they matter.
Prototype
Free
$0/mo
Start for freeNo card required. Upgrade any time from the dashboard.
Build against the same production pipeline the paid tiers use. Zero card, zero commitment.
- 10,000 credits every month
- Full 12-voice preset catalogue · Playground
- Community support
Ship it
Creator
$19/mo
The anchor plan for individual developers and small teams going to production.
- 500,000 credits every month
- All presets plus 30 custom voices · overage billing
- Email support, 24-hour target response
Custom
Enterprise
Custom
Usage-based pricing, signed MSA, data-processing addenda, and the residency posture your legal team asked for.
- Signed DPA and MSA · custom retention · BAA on request
- Pick any of 4 live regions · dedicated tenants on contract
- Unlimited voices · unlimited keys · on-prem optional
- Slack Connect · dedicated CSM · 24/7 sev-1
One more thing
Your users are waiting for a voice that sounds like them.
Five thousand free characters today. Twelve voices and every dialect we support. Thirty seconds to your first synthesis.
No card. Upgrade when, and if, you grow out of the free tier.