Trained on someone else’s Arabic
Mainstream models learn mostly from formal text and from Egyptian, Levantine and Gulf speech. Everyday Yemeni words, names and places come back garbled.
Private beta — early access open
AI that understands how Yemen actually speaks.
YemenAI builds speech and language models for Sanaani, Adeni, Taizi, Hadrami and Tihami Arabic — the dialects mainstream models mishear. Put a dialect-fluent assistant on your WhatsApp, or build on our API.
- Voice notes and text
- Five Yemeni dialects
- Human handoff built in
Exchange shop · WhatsApp
Demo
0:05
Transcript
ar-YE
يا خبير، اشتي أحوّل مية ألف لأمي في صنعاء ذلحين. كم الصرف اليوم؟
English “Friend, I want to send 100,000 to my mother in Sana’a now. What’s today’s rate?”
Dialect · Sanaani
Intent · Transfer + rate
حيّاك يا خبير! أرسل لي اسم المستلم ورقم تلفونه، وأنا أرسل لك سعر اليوم وأجهّز الحوالة.
The problem
Yemen talks in voice notes. Most AI isn’t listening.
Arabic AI has improved fast — for Modern Standard Arabic and the largest dialects. Yemeni speech sits at the edge of that data, and it shows the moment a real customer presses record.
Voice-first, not form-first
Customers send voice notes, not typed forms. Staff spend hours listening, replying and re-typing orders by hand.
One country, many dialects
A pharmacist in Mukalla and a remittance agent in Taiz hear different Arabic. A single “Arabic” setting serves neither of them well.
- 30M+
- people speak Yemeni Arabic inside Yemen
- Voice-first
- voice notes are the everyday way to message
- Gulf-wide
- a large diaspora lives in Saudi Arabia and across the Gulf
Market context based on public estimates. Not company metrics.
Products
One dialect engine. Three ways to use it.
Everything we build runs on the same Yemeni speech and language models, tuned to the way people actually talk.
For businesses
YemenAI Assistant
A WhatsApp and voice-note assistant that answers customers in their own dialect, takes orders and bookings, and hands the conversation to your team when a person should step in.
- Understands voice notes and text
- Quotes, orders, bookings and FAQs
- Handoff to staff with a written summary
- Replies by voice or text, in dialect
For developers & enterprises
YemenAI API
Dialect speech-to-text, natural text-to-speech and dialect-aware language endpoints — for telcos, banks and NGOs running hotlines, and for teams building their own products.
- Speech-to-text with dialect tags
- Yemeni text-to-speech voices
- Dialect-aware chat, summaries and classification
- Streaming and batch
For the community
Data & Research
A consented, community-sourced corpus of Yemeni dialect voice notes. Contributors are paid for every clip that passes verification.
- Explicit, revocable consent
- Paid per verified clip
- Dialect, region and speaker metadata
- Licensing for research partners
How it works
From voice note to done, without retyping a word.
The Assistant is designed to be set up by the business owner, not an IT department.
-
Connect your number
Link your WhatsApp Business number through the official platform.
-
Teach it your business
Add prices, products, hours and rules — in Arabic, by text or by voice.
-
It answers in dialect
Customers get quick, accurate replies in the dialect they wrote or spoke in, day and night.
-
Your team steps in
Anything sensitive or unusual goes to a person, with a summary, so no customer has to repeat themselves.
Use cases
Built for the businesses Yemen runs on.
The conversations are familiar. Now they can be answered at any hour, in the customer’s own words.
Exchange & remittance
«كم الصرف اليوم؟ اشتي أحوّل لعدن»
Rate quotes, transfer details and ID checklists, then a clean handoff to complete the transaction.
Pharmacies
«بِه عندكم هذا الدواء؟ صوّرت لك الروشتة»
Stock and price questions, alternatives referred to the pharmacist, delivery or pickup arranged.
Telecom resellers
«اشتي باقة نت لهذا الرقم»
Bundle and top-up requests captured accurately, with the right number and the right amount.
Transport operators
«بِه باص بكرة من صنعاء لعدن؟ كم الكرسي؟»
Schedules, seat availability, fares and bookings across every route you run.
Telcos & banks
«رصيدي انخصم وما عرفت ليش»
Call-centre audio transcribed, tagged and routed by dialect and intent, with summaries for agents.
NGOs & hotlines
«متى توزيع السلال الغذائية في منطقتنا؟»
Feedback lines understood at scale, urgent cases triaged and trends reported, under strict privacy.
Dialect coverage
Five dialects, one model family.
We model Yemeni Arabic as it is spoken, region by region, and tag every transcript with the dialect it heard.
Also understood Modern Standard Arabic and Gulf Arabic, for customers in the diaspora.
Research roadmap Mehri and Soqotri, developed with community partners.
For developers
Yemeni speech, one API call away.
Send a WhatsApp voice note as it is. Get back the text, the dialect it was spoken in, and word-level timestamps.
/v1/speech/transcribeDialect speech-to-text/v1/speech/synthesizeYemeni text-to-speech/v1/chatDialect-aware language model/v1/dialect/identifyDialect identification
# Transcribe a WhatsApp voice note
curl https://api.yemenai.com/v1/speech/transcribe \
-H "Authorization: Bearer $YEMENAI_API_KEY" \
-F "audio=@voice-note.ogg" \
-F "dialect=auto" \
-F "timestamps=word"const form = new FormData();
form.append("audio", voiceNote); // .ogg / .opus
form.append("dialect", "auto");
const res = await fetch(
"https://api.yemenai.com/v1/speech/transcribe",
{
method: "POST",
headers: { Authorization: `Bearer ${key}` },
body: form,
}
);
const { text, dialect } = await res.json();{
"text": "يا خبير، اشتي أحوّل مية ألف لأمي في صنعاء ذلحين",
"language": "ar-YE",
"dialect": {
"label": "sanaani",
"confidence": 0.94
},
"duration_s": 4.8,
"words": [
{ "w": "اشتي", "start": 0.62, "end": 0.98 },
…
]
}Preview of the planned API. Endpoints, fields and values are illustrative and may change before general availability.
Contribute to the corpus
Your voice teaches the model. You get paid for it.
The best Yemeni dialect data will come from Yemenis, at home and abroad, on fair terms. We are building that corpus openly, with consent at every step.
Join
Tell us your dialect and where you grew up.
Record
Read short prompts, or answer everyday questions in your own words.
Verify
Another native speaker checks each clip for quality and dialect.
Get paid
You are paid for every clip that passes.
Consent first
Clear terms in plain Arabic, and you can withdraw your clips.
Private by default
Names, numbers and personal details are removed before any clip is used.
Fair pay
A published rate for every verified clip, paid where you live.
Early access
Join the private beta.
We are onboarding a small number of businesses, developers and contributors at a time. Tell us a little about yourself and we will be in touch.
What happens next
- Your email app opens with your details filled in.
- You send it to hello@yemenai.com.
- We reply with next steps for your use case.
The team
The people building YemenAI.
YemenAI is built and run by the Basha Investments group team. These are the people who work on it.
-
Osama Basha
Chairman
-
Ghaith Almasri
Head of Operations
-
Afaf Shawwa
Language Data Program Lead
-
Yasser Basha
Head of Partnerships
-
Faysal Kurbeh
Enterprise Partnerships
-
Martina Medhat
Developer Relations
-
Nour Almohamade
Enterprise Sales
-
Yosef Elwardany
Finance Lead
-
Fady Essam
Talent & Contributor Network
-
Mohammed Abbas
Security & Data Protection
-
Mohammed Fakhrani
Lead Engineer
-
Mennatullah Assem
Product Designer
-
Farida Dahshan
Growth Marketing
Which dialects do you support?
Our models focus on Sanaani, Adeni, Taizi, Hadrami and Tihami Arabic, alongside Modern Standard Arabic and Gulf Arabic for customers in the diaspora. Every transcript is tagged with the dialect it detected.
Can I use YemenAI today?
We are in private beta and onboarding partners in small groups. Request early access and tell us about your use case. We prioritize teams for whom dialect accuracy matters most.
How does the Assistant connect to WhatsApp?
Through the official WhatsApp Business Platform, using your business number. Your customers keep messaging you exactly as they do today.
What happens to my customers’ messages?
They are processed to serve your business and nothing else. Customer conversations are not used to train shared models unless you opt in, and you can ask us to delete them at any time.
How will pricing work?
Beta partners receive pricing during onboarding. The Assistant is planned as a monthly subscription, and the API as usage-based.
I live outside Yemen. Can I contribute recordings?
Yes. Native speakers in Saudi Arabia, the Gulf and beyond are welcome. Payment options depend on your country, and we confirm them before you record anything.
Who is behind YemenAI?
YemenAI is part of the Basha Investments portfolio.