The enterprise’s choice

The voice layer for the Arab world.

0%

STT accuracy in real-life conditions

+0

Arabic dialects, MSA, and English language support.

0+

On-prem solutions for enterprises across MENA.

<0ms

Average response latency for voice agents

The enterprise’s choice · in production at

Try it

Hamsa Media,
in one frame.

Studio tools for speech in both directions, and a library of voices to hear them in. Pick one and press play. Nothing needs signing up for.

Turns written Arabic into speech that sounds like a person reading it.

Up to 300 characters

Text to Speech · text in → speech out

No sign-up
77/300

Type anything, pick a dialect and a voice, then press play.

Free to start · in your browserMake voiceovers in Hamsa Media

Industries

Wherever the Arab world talks.

All industries→

Proprietary AI models

Our own models.
Built for how the region speaks.

Hamsa doesn’t wrap someone else’s API. Two foundation speech models, trained in-house on proprietary, highly-supervised data.

Hear both models→

AI Model · Speech to text

#1 of 5 · 2026

Hamsa-STT-S4

model="hamsa-stt-s4"

96%

overall accuracy across 13+ dialects, on real-life audio

<100ms

streaming latency, built for real-time conversation

speech in → text outkhaleeji · S1 · M
partial…

بدّيأعرفالأسعارلوسمحت

✓ numbers & emails✓ no-speech detection✓ backchannels✓ code-switch AR⇄EN✓ speaker ID · S1/S2✓ gender M/F✓ timestamps

streaming · batch · real-time

Explore the model →

AI Model · Text to speech

omni · v3 · 24kHz

Hamsa-TTS-Omni v2.8

model="hamsa-tts-omni-v2.8"

13+

Arabic dialects, plus English in all accents

<150ms

streaming latency to the first byte of synthesized audio

text in → speech outvoice: layla · 24kHz
speaking…

مرحباً!كيفأقدرأساعدك؟

✓ expressive✓ natural prosody✓ human-like✓ emotion & tone✓ voice cloning

cloning & custom voices · enterprise

Explore the model →

Also in the stack: the models around the models

hamsa-turn

Semantic Turn Detection

Knows when the speaker is actually done: by meaning, not by silence.

hamsa-turnsilence ≠ done
hamsa-bc

Interruption & Backchannels

«تمام» and «aha» are listening signals. Real interruptions still get through.

hamsa-bclistening vs interrupting
hamsa-gender

Gender Detection

Male / Female tags on every speaker, for routing and analytics.

hamsa-genderM / F on every speaker

For developers

Five lines to
voice-enable your app.

One JavaScript SDK for voice and chat agents. Streaming transcription, synthesis and voice cloning over REST and WebSocket. Dialect-aware, production-ready.

import { HamsaVoiceAgent } from '@hamsa-ai/voice-agents-sdk';

const agent = new HamsaVoiceAgent({ publicKey });

agent.on('answerReceived', (text) => console.log(text));
await agent.start({ agentId, voiceEnablement: true });
// → "مرحباً! أنا ليلى من همسة"
REST APIWebSocket streamingVoice Agents Web SDK

The enterprise’s choice

Why enterprises pick
Hamsa.

From data sovereignty and guardrails to the engineers who sit with your team through go-live: the practical reasons banks, telecoms, and government platforms across the region run Arabic voice AI on Hamsa.

On-premise and data sovereignty

01

Run the full stack inside your own infrastructure: private cloud or fully air-gapped on-premise. The models come to your data instead of your audio going to someone else’s cloud.

Hamsa cloud

managed, in-region

✓

Private cloud

your own tenancy

✓

On-premise

air-gapped, your racks

✓

Guardrails tuned for production

02

Flow agents keep the conversation on a graph you can see: answers grounded in your knowledge base, tools called only where you allow them, and a human handoff the moment the agent is out of its depth.

Grounded answers

from your documents only

Scoped tools

per node, never open-ended

Escalation path

human handoff, full context

built in

Implementation services, not a docs link

03

Solution engineers design the flows, wire your systems, and sit with your team through go-live. It’s the difference between a pilot that demos well and an agent that survives real call volume.

01

Discovery

intents, systems, success metrics

02

Flow design

built with your team, in dialect

03

Integration

CRM, telephony, knowledge base

04

Go-live

tuning against real calls

On-ground support across MENA

04

Teams in the region, in your time zone, working in Arabic. Procurement reviews, security questionnaires, and production incidents are handled by people who answer during your working hours.

Same time zone

no overnight ticket queues

Arabic-first support

in the language you work in

Named engineer

through pilot and production

enterprise plan

Models owned end-to-end

05

Speech-to-text and text-to-speech are trained in-house. No third-party AI provider in the loop means nothing trains on your calls, and model quality is never someone else’s roadmap.

Caller audio

encrypted in transit

Hamsa models

trained in-house

no third party

Your systems

transcripts & outcomes

Data residency, in region

06

Calls, transcripts, and recordings stay inside the region your regulator requires, with on-premise available where nothing may leave the building at all.

Region

KSAUAEQatarJordanEUOn-prem

Audio never leaves the selected region.

Access control and audit trails

07

Separate projects and roles per environment, credentials stored once and referenced by name, and every call retained with its transcript and recording, filterable and exportable for QA.

Projects & roles

per environment, public & secret keys

Secrets store

referenced by name, never pasted

Call history

transcripts & recordings

export ↓

Deployment & sovereignty

Your voice AI.
On your terms.

The same models and APIs, wherever your data is allowed to live, from our cloud to hardware you can physically touch.

01

Hamsa Cloud · SaaS

Start in minutes. Encrypted in transit and at rest, with no third-party AI providers anywhere in the chain.

✓ Zero infrastructure✓ Running in a day
02

Private Cloud

The full Hamsa stack deployed inside your own VPC or dedicated tenancy: your cloud account, your encryption keys, your network rules. Nothing leaves your perimeter.

✓ Your VPC & keys✓ Dedicated tenancy✓ No public internet
03

On-premise

Total sovereignty. Models, storage, and logs run on your own metal: air-gap ready, zero external calls. Built for governments, banks, and defense-grade compliance.

✓ Air-gap ready✓ Zero external calls✓ Full sovereignty

Architecture · Hamsa Cloud

SaaS
▤

Your app

Web & mobile

Telephony / SIP

audio in · TLS 1.3

response · <300ms

Hamsa Cloud

STT-S4TTS-OmniAgentsStorage

AES-256 at rest · no 3rd-party AI

Architecture · Private cloud

Your VPC
◫

Your users

internal apps & lines

private link

Your VPC · your cloud account

no public internet

Hamsa stack · inside your tenancy

STT-S4TTS-OmniAgents
your KMS keysyour IAMyour audit logs

Architecture · On-premise

air-gapped

Your data center

sovereign

GPU node · model weights

STT-S4 · TTS-Omni

Agents runtime

flows · tools · logs

Storage

audio · transcripts · audit

weights · audio · transcripts: never leave this box

External cloud

any provider

⊘ zero calls out

Start in minutes. Encrypted in transit and at rest, with no third-party AI providers anywhere in the chain.

✓ Zero infrastructure✓ Running in a day✓ Encrypted end-to-end✓ Data stays in-region✓ SLA 99.9%

Partner ecosystem

Trusted Across the
Technology Ecosystem

Hamsa works with leading technology companies, integrators, and solution providers to bring enterprise Voice AI to organizations across the region.

FAQ

Questions,
answered.

Which Arabic dialects does Hamsa support?+

Hamsa STT supports 13+ Arabic dialects, including Standard Arabic, Egyptian, Gulf, Levantine and Iraqi, and handles Arabic–English code-switching within a single sentence. Hamsa TTS speaks the same dialects, plus English in all accents.

How do I integrate Hamsa into my product?+

Use the REST API for transcription, synthesis and voice cloning, or the Voice Agents Web SDK to put an agent in your app. Agents also deploy to phone lines; most teams ship a first integration in under a day.

Can Hamsa run on-premise?+

Yes. Enterprise plans include private cloud and on-premise deployment with the same models and APIs, for teams with data-residency requirements.

How is pricing structured?+

Plans start free with 50 credits per month and scale by usage: voice-agent minutes, transcription hours, and synthesis characters all draw from one credit pool.