Live Translate
Company-governed translation for frontline conversations — runs through your own provider or MangoApps' managed one, no conversation history kept.
Overview
Live Translate shares the company glossary MangoApps applies to policy, alert, Sites, and voice-note translation — the list your team curates once (brand names, equipment, SKUs) enforces on every spoken turn here too, not a separate list per tool.
Translation runs through your own provider when an admin connects one, or MangoApps' managed translation service otherwise, and speech runs on MangoApps' OpenAI account — never an outside consumer app your compliance team can't see into. Each turn is transcribed, translated, and spoken back in real time. No conversation history is kept: a one-device turn isn't saved, a two-device turn passes through MangoApps' message service on its way to the other phone and is deleted within about 90 minutes, and a phrase someone flags as wrong is saved for admin review.
Two people share one phone, or each joins from their own with a short session code — and on a paired session, hands-free mode keeps translating as you speak, no button required. The AI handles the mechanics — hearing, translating, speaking back — so two people who don't share a language can talk directly, without a bilingual colleague relaying between them; the conversation stays theirs.
Highlights
Capabilities
Company-governed & private
-
Runs through your organization's own translation provider (bring-your-own AWS Translate or DeepL)
-
Managed AWS Translate default when no provider is configured
-
Works with zero setup — connecting your own provider is optional
-
Ephemeral sessions — no conversation history is kept Two qualifications: a turn you flag "Wrong?" is saved for admin review, and in paired two-device mode each turn travels to the other phone over MangoApps' shared real-time message service, labelled with your organization's ID, and is deleted within roughly 90 minutes (one-device mode doesn't use it).
-
Names the active provider in-app so users see who is translating Only shown once you've configured your own BYOK provider — the managed default stays unnamed.
Speak & hear (turn-based interpreter)
-
Push-to-talk, turn-based speech-to-speech interpretation
-
Spoken playback in the listener's language (not just text)
-
On-screen transcript of the exchange (source + translation) In two-device mode both phones show each turn's words and translation; only the listener's phone speaks it.
-
Automatic speaker-language detection ("Detect language") in one-device mode
-
Hands-free continuous mode — speak without holding the button (two-device pairs)
One device or two
-
Single-device mode (two people share one phone)
-
Paired two-device mode over a short session code
-
Code-gated pairing — only the two who know the code can join or inject turns
-
Remembers each side's language across visits
Company glossary
-
Company terminology enforced provider-independently (protect → translate → restore)
-
Term mappings — a matched term renders as your approved target A mapping applies to turns in its exact language pair, such as English → Spanish.
-
Do-not-translate terms — brands, SKUs, and codes pass through untouched
-
Case-sensitive term matching option
-
Glossary managed by admins at /admin/translation
Safeguards & metering
-
Per-user hourly rate limits on turns and session creation
-
Admin-controlled audience — publish to everyone or to chosen groups (admins and app-role holders always have access)
-
Every turn's AI usage logged for cost tracking
Limits & Specs
-
Languages: 15, plus automatic Detect of the speaker's language in one-device mode
-
Turns per user, per hour: 200
-
Paired sessions per hour: 60
-
Max recording per turn: 2 MB (short push-to-talk turns)
-
Translation engines: AWS Translate (managed, or your own) or DeepL (your own)
-
Speech in / out: OpenAI Whisper (transcribe) / OpenAI TTS (speak)
-
Pricing: Pay-as-you-go — no license; Whisper, translation, and TTS meter through your AI/Voice credit pool
Use cases
FAQ
No conversation record is kept — each turn is transcribed, translated, and spoken back in real time, and there is no transcript history to open afterwards. Two qualifications, both worth knowing if you are answering a compliance question. Tapping "Wrong?" on a turn deliberately saves that phrase and its translation to an admin review queue so your team can correct the glossary — that is the only durable copy. And in two-device mode each turn travels to the other phone over MangoApps' shared real-time message service, labelled with your organization's ID, and is deleted within roughly 90 minutes; until then it holds the turn's text and audio, no screen in the product reads it, and one-device mode never uses it.
It uses your organization's own translation provider — a dedicated machine-translation engine such as AWS Translate or DeepL — when an admin connects one; otherwise MangoApps' managed translation service (AWS Translate on MangoApps' account). A spoken turn never switches engines when a provider fails — the turn fails and the speaker tries again. It is the same provider chat, recognition, and voicenote translation use.
It is turn-based: each speaker holds to talk, or — on a paired two-device session — switches on hands-free, which listens continuously and sends each utterance when the speaker pauses. Either way, each utterance is translated as its own turn; there's no simultaneous, talk-over-each-other calling.
Pay-as-you-go. There is no license to buy — every tenant can turn it on, and the Whisper, machine-translation, and text-to-speech usage each turn consumes is metered through your existing AI/Voice credit pool.
Yes. Your company glossary — both term mappings and do-not-translate rules — is applied on every turn, provider-independently, and is managed by admins at /admin/translation.
Fifteen languages out of the box — English, Spanish, French, German, Portuguese, Italian, Chinese, Japanese, Korean, Arabic, Hindi, Russian, Vietnamese, Tagalog, and Polish — plus, in one-device mode, a "Detect language" option that recognizes a speaker's language from their speech, so it doesn't have to be picked ahead of time. The side set to Detect has to speak first, so the app knows which language to translate the reply into.
No. If your organization sets up its own provider (AWS Translate or DeepL), Live Translate runs through it; if none is configured, it uses a managed AWS Translate default, so it works with zero setup. Speech-to-text runs on the platform's OpenAI connection, so it must be available for a turn to run; if text-to-speech fails, the turn still arrives as on-screen text.
Yes. Beyond single-device mode (two people sharing one phone), one person can create a paired session and share a short code; the other joins from their own phone and hears each turn spoken in their own language. Whoever enters the code first claims that second seat, and only that pair can inject turns — a third person who happens to have the code cannot join or add audio to the conversation. Live Translate isn't in the MangoApps mobile app — on a phone, open it in the phone's web browser: sign in at your company's MangoApps address followed by /m/login, then open Live Translate from "More" (or go to /apps/live-translate).
No. Live Translate is a spoken interpreter — you hold a button to talk and the other person hears the translation played back. There is no type-to-translate text box; it is speech-in, speech-out only.
They are separate apps for different needs. Live Assist is real-time "see what I see" video help — a remote expert joining a live video/screen-share session launched from a work object. Live Translate is a spoken language interpreter for clearing a language barrier in a face-to-face or two-phone conversation. Use Live Assist when someone needs to show a problem on camera; use Live Translate when two people need to understand each other across languages.
