Feature
Transcription that speaks your Arabic.
32 Arabic dialects, 18 English dialects, and the code-switching in between. Every line attributed to a named speaker, with your organization's terms kept exact.
Built Arabic-first
Real meetings do not speak textbook Arabic.
A Gulf accent opens the meeting, a Levantine aside answers it, and the numbers land in English mid-sentence. Generic speech models guess their way through that. MeetriX was engineered for it from the first line of code.
32 Arabic dialects
From Khaleeji to Maghrebi, one engine.
The MeetriX speech engine understands 32 Arabic dialects plus Modern Standard Arabic, and 18 English dialects beside them. They are the ones your meetings are actually held in, from Gulf and Egyptian to Levantine, Iraqi, and Maghrebi.
- Gulf, Egyptian, Levantine, Iraqi, Sudanese, Maghrebi and more
- Modern Standard Arabic for formal sessions and media
- 1.62% character error rate on Arabic meetings and 1.83% on English meetings in Lisan's September 2026 evaluation, lower than all six external ASR configurations compared
The right-hand column summarises typical cloud speech-to-text products; providers change, so verify against your own audio. The measured comparison is on the M3 results page.
Code-switching
When the sentence changes language, MeetriX keeps up.
"أرسلوا اتفاقية SLA مع العرض المحدّث" is one sentence in two languages, and it is how the region does business. In Lisan's September 2026 evaluation on real Arabic meetings, M3 kept English technical and business terms inside Arabic speech more consistently than the other systems compared; several of them dropped the embedded term or wrote it phonetically.
- English terms inside Arabic sentences transcribed intact
- Each line tagged with its language, readable in both directions
- Translation to a single language afterwards, if you want it
يجب أن يشمل السعر النهائي الصيانة، لا التركيب فقط.
We can include maintenance for year one, with an SLA attached.
حسنًا، أرسلوا اتفاقية مستوى الخدمة مع العرض المحدّث قبل الخميس.
I will confirm the commercial terms on our side by Wednesday.
Named speakers
"Speaker 1" is not a record. Names are.
MeetriX separates voices and then resolves them to the actual participants in the meeting, so the transcript reads Maha said, Omar answered, not Speaker 1 and Speaker 2. Minutes, per-speaker reports, and talk-time analytics all inherit real names.
- Voice separation plus participant matching for real names
- Per-speaker reports: every attendee can get their own recap
- Talk-time analytics built on the same attribution
Talk time, per-speaker reports, and minutes all inherit the same named attribution.
Custom vocabulary
Your names, products, and terms, exactly right.
Add your organization's canonical terms once: product names, people, departments, project codenames. MeetriX keeps them exact across transcripts and summaries, including the tricky part nobody else handles: Arabic and English transliterations of the same name staying consistent.
- Organization-level dictionary of canonical terms
- Arabic ↔ English transliterations normalized to one spelling
- Applied automatically to every meeting in the workspace
Terms hold their exact spelling across every transcript and summary.
Under the hood, honestly
Our engine, our benchmarks, your audio.
The speech engine is built and hosted by Lisan, not rented from a cloud API, which is also why it can run fully on your own servers. The numbers on this page come from Lisan's September 2026 evaluation on real Arabic meetings. The benchmark that matters is your audio: start free and test it on a real meeting today.
Measured, not promised
Arabic and English, the same rules for every system.
In Lisan's September 2026 evaluation, M3, the system behind this page, recorded a 1.62% character error rate on real multi-dialect Arabic meetings and 1.83% on real English meetings, both with human-verified transcripts, lower than each of the six external ASR model configurations compared. The scoring rules, nine transcript examples and the cases where M3 itself slips are published with the results.
The M3 evaluation report →Arabic meetings
Character error rate (CER). Lower is better.
English meetings
Character error rate (CER). Lower is better.
Lisan evaluation, September 2026: the same audio, human-verified reference and scoring rules for every system within each language. The two languages use separate evaluation sets, so compare systems within a chart, not across the two.
Every source of audio
Live meetings on four platforms, browser recordings, and uploaded files all run through the same engine. In-person meetings →
Search what was said
Transcripts are searchable, so "what did we agree with the vendor in May" is a query, not an hour of scrubbing audio. See the tour →
Never trained on your data
Your meetings improve your record, not our models. On-prem deployments keep the audio inside your network entirely. Security →
Frequently asked questions
Which Arabic dialects does MeetriX understand?
32 dialects across the Arab world, from Gulf and Egyptian to Levantine, Iraqi, and Maghrebi, plus Modern Standard Arabic.
How accurate is it?
In Lisan's September 2026 evaluation on real business meetings with human-verified references, M3 reached a 1.62% character error rate on multi-dialect Arabic meetings (98.38% character accuracy) and 1.83% on English meetings (98.17%), lower than each of the six external ASR model configurations compared. The scoring rules and examples are published at the results page, and the real test is a pilot on your own meetings.
What happens when people mix Arabic and English?
That is a headline capability, not an edge case. English terms inside Arabic sentences (and the reverse) are transcribed intact, and in Lisan's September 2026 evaluation on real Arabic meetings M3 kept English terms inside Arabic speech more consistently than the other systems compared.
How does it know who is speaking?
The engine separates voices, then matches them to the actual meeting participants, so lines carry real names instead of Speaker 1. Those names flow into minutes, per-speaker reports, and talk-time analytics.
Can I teach it our product and people names?
Yes. The custom dictionary stores your canonical terms and keeps them exact in every transcript and summary, including consistent Arabic and English transliterations of the same name. See teams & admin.
Can the transcript be translated?
Yes, transcripts and summaries translate between Arabic and English, so a bilingual meeting can produce a single-language record for each audience. See translation.
Stay in the conversation. MeetriX takes the notes.
It joins the call, transcribes who said what in Arabic or English, and sends the summary and action items before you are back at your desk. 600 free minutes when you connect your calendar.
Arabic & English · 32 Arabic dialects · No credit card required