Live speech translation · Built by Ordinexis
Say it once.
They hear it in their own tongue.
One person speaks. Everyone in the room — or anywhere in the world — hears the translation spoken aloud, live, with subtitles on screen. No app to install, no booth to build, no interpreter to book days in advance.
Direct-to-provider audio path Two proprietary AI speech models Up to 100 languages Built for Malaysia & Asia Pacific's multilingual rooms
one voice → every language
How it works
Three steps. Nothing to install.
Because this genuinely is a sequence — pick a source, pick a language, start talking — it's the one place on this page numbers earn their keep.
What are you translating?
A headset mic, a room mic, a mixer line-in, or the audio from a browser tab or app — even a mic and a tab mixed together with a live balance fader and a limiter so nothing clips.
Translate into
Pick up to five output languages at once, from a curated set of 13 that speak aloud in a steady voice, or a reach of 100 for text and regional-accent control. Each output gets its own volume, mute, solo and audio device.
Start
Audio streams straight from the browser to the AI speech model — subtitles appear immediately, spoken translation follows a beat behind. Auto-stop after silence means a forgotten tab doesn't quietly keep billing.
Your browser
Microphone or tab audioAI speech model
One of two engines, directWhere it earns its keep
A hall, a tour bus, a clinic, a broadcast — one platform.
Anywhere one voice needs to reach people who don't all share a language. That describes most of Asia Pacific, most days of the week.
Live captioning + spoken translation
Worship services & multi-faith congregations
Malaysia's mosques, churches and temples often serve congregations who arrive fluent in Bahasa Melayu, Mandarin, Tamil and English — in the same room, the same service. One sermon, spoken once, becomes up to five spoken languages plus calm, projector-friendly captions on the hall's screen.
Up to 5 simultaneous outputs
Conferences, summits & trade missions
ASEAN-style gatherings in Kuala Lumpur, Singapore, Jakarta and Bangkok often book a wall of interpreter booths for one afternoon. Salam.AI replaces the booth with a browser tab — five language channels out at once, each with its own volume, mute and output device.
Tab or mic, phone-friendly listening
Guided tours & attractions
A guide at Batu Caves, the Petronas Towers or a Langkawi jetty says one thing once — a coach of visitors from China, Korea, India, the Gulf and Europe each hears it in their own language, on their own phone, over a link with nothing to download.
Live captions for every seat
Classrooms, lectures & campus life
Malaysia's universities and international schools already teach into classrooms where English, Bahasa Melayu and Mandarin sit side by side. A lecture spoken once becomes live subtitles — and, where useful, spoken audio in a student's stronger language.
Two-way, push-to-talk
Clinics, embassies & consulates
A patient speaking Bahasa Melayu, Mandarin, Tamil or Burmese and a clinician who only speaks English can hold a real back-and-forth on one shared microphone — hold to talk, release, hear the reply. No interpreter to book, drive in or wait for.
One link, unlimited listeners
Corporate town halls & regional broadcasts
One all-hands recorded once in Kuala Lumpur or Singapore fans out — live captioned and spoken — to teams in Jakarta, Manila, Bangkok, Hanoi and Tokyo, at the same cost whether ten people tune in or ten thousand.
Malaysia's own multilingual reality
Government & public communication
Public briefings and civil-service counters serving a nation that runs daily on Bahasa Malaysia, English, Mandarin and Tamil — plus a large resident workforce speaking Bahasa Indonesia, Burmese, Nepali and Bengali — reach every group from a single spoken announcement.
Every ticket, every language, no queue
Customer support & contact centres
A support line in Kuala Lumpur, Manila or Ho Chi Minh City can pick up a call in one of a hundred languages without holding for a specialist agent or a callback queue.
Regional focus
Built for the region's real linguistic map
Malaysia alone runs on four languages in daily public life — Bahasa Malaysia, English, Mandarin and Tamil — plus Cantonese and Hokkien in commerce, Iban and Kadazan-Dusun across Sabah and Sarawak, and a resident migrant workforce speaking Bahasa Indonesia, Burmese, Nepali, Bengali and Tagalog. That's before a single tourist, delegate or student walks in. Salam.AI's 100-language reach, grouped by region for Southeast Asia, East Asia, South Asia and beyond, was shaped by exactly this kind of room — one where five languages are already speaking to each other before anyone says a word.
Languages
One voice. A hundred rooms.
Two engines, two honest numbers — and a live board that never stops finding another language to flip to.
13 languages speak aloud today with a curated, steady voice — English, Indonesian, Vietnamese, Chinese, Japanese, Korean, Hindi, Spanish, Portuguese, French, German, Italian and Russian. 100 languages can be reached through the broader engine, with regional accent control (en-GB, pt-PT, cmn-TW, ms-MY) and an optional gendered voice.
Ask the curated engine for a language it doesn't officially cover and it quietly falls back to its nearest match rather than failing outright — which is why its own menu stays pinned to exactly the 13 it validates.
↑ a rotating sample of the full 100-language list, always in motion
Featured
Southeast Asia — 9 languages
East Asia — 4 languages
South Asia — 4 languages
Europe — 8 languages, including
Middle East — 3 languages, including
71 more, A–Z
Reach an audience
One link. A hundred rooms just got bigger.
Turn any live translated session into a shareable listen link. One shared stream fans out to any number of listeners — it is not one paid session per person, so audience size never multiplies the bill.
- → A big tap-to-play listener page — no login, no app, works on Safari/iOS and every other browser.
- → Roughly 6–20 seconds of delivery delay — the honest trade-off for reach that scales to any crowd, at constant marginal cost.
- → Broadcast start/stop stays authenticated and scoped to your own session; the public page only ever needs the link.
Security & trust
What it never sees is the point.
The server holds control-plane data — auth, limits, metadata. It never holds plaintext audio, and never holds a long-lived provider key.
Two-step sign-in, invite-only teams
Email + password, plus TOTP authenticator MFA (any standard app) with one-time backup codes. Admins invite teammates by email; nobody but the invitee ever sets or sees their password.
Full tenant isolation & audit log
Every tenant's sessions, recordings and usage are scoped and invisible to every other tenant. Logins, MFA events, admin changes, broadcast start/stop and recording purges all land in an audit trail.
Client-side, end-to-end encrypted recordings
Recording is one click, optional. Audio is encrypted in the browser (AES-256-GCM) before upload — the server only ever stores ciphertext, a SHA-256 integrity manifest and a wrapped content key. Playback re-verifies the hash and decrypts locally.
3f9ae2·AES-256-GCM·b71cd0·SHA-256 ✓
Short-lived, single-purpose credentials
AI provider API keys never reach the browser. The server mints a scoped, short-lived client secret per session — and every budget or quota check is enforced atomically, server-side, before a paid session is ever minted.
For teams & organizations
Pay for minutes translated — not seats, not listeners.
Cost is metered by audio minutes, not headcount. A usage dashboard and admin-configurable budgets keep it predictable.
A cost alert fires automatically at ~80% of the daily minute budget — before it becomes a surprise invoice.
Three roles, one tenant boundary
Every tenant's data, sessions and recordings stay invisible to every other tenant — role changes just decide who inside yours can do what.
Member
Runs sessions, shares listen links, records with permission.
Team Admin
Invites & disables teammates, reads the usage dashboard.
Super Admin
Creates tenants, sets limits and retention platform-wide.
The honest comparison
What it actually costs to translate a room, today
Not to replace every human interpreter — a seasoned conference or court interpreter still does a different, harder job. This is about the far more common case: a meeting, service, tour or broadcast that today gets either no translation, or one booked at real cost and real notice.
typical booking lead time for a qualified interpreter, longer for less common pairs like Malay ↔ Khmer or Malay ↔ Burmese
from opening a browser tab to a live, spoken, subtitled translation — any hour, any day
Cost to reach more people
Conceptual — illustrates how the two cost structures diverge as an audience grows. Not literal pricing.
← swipe to see the full comparison →
| Dimension | A professional human interpreter | Salam.AI |
|---|---|---|
| Time to get started | Booked days to weeks ahead; last-minute requests often go unfilled for less common pairs. | Open a tab and start in seconds, any hour, any day. |
| Languages per session | One language pair per interpreter booked — a 5-language event needs 5+ separate bookings. | Up to 5 simultaneous output languages from one session, 100 to choose from. |
| Scaling to a bigger audience | Bounded by the room, the headset inventory and the interpreter's stamina; a second site means booking again from scratch. | One listen link fans out over HLS to any number of listeners at constant marginal cost. |
| Session length & fatigue | Simultaneous interpreting is demanding — industry guidance calls for a second interpreter and rotation roughly every 20–30 minutes. | Runs the full session — a two-hour service or a six-hour conference — without a rotation. |
| Travel & equipment | Travel, accommodation and booth/headset hire for on-site work; remote interpreting still needs dedicated gear. | A browser and a microphone — the room's own laptop or a phone tab is enough. |
| Recordings | Rarely recorded with a documented chain of custody by default. | Optional one-click recording, encrypted end-to-end in the browser before upload. |
| What drives the cost | Per interpreter, per hour or per day, with half/full-day minimums, multiplied by every language and every location. | Metered by minutes of audio translated — one predictable unit, whichever of the 100 languages is chosen. |