Frequently asked questions about AI Translator

AI Translator is a desktop app that shows live translated subtitles for any audio playing on your computer, such as a meeting, webinar or video. It recognizes speech and translates it on your own machine, then shows the translation on a floating subtitle bar. See the full list of features.

Yes. The app captures system audio, so it does not depend on the meeting app and needs no per-app setup: you just press Start. On macOS you can also choose to listen to a single app only (Settings › Audio) so notification sounds are not translated.

We have checked system-audio capture on macOS 26 with Zoom, Google Meet (Chrome, Safari, Edge), Microsoft Teams and Zalo PC: the speaker's audio was captured clearly. That was a check of the capture component, confirmed by the tester, not a formal acceptance test on the release build, and we have not tried every macOS version or every app. These names are used only to describe compatibility, and AI Translator is not affiliated with them. See meeting translation.

None of the three. No bot joins your meeting, no plugin goes into your meeting app, and there is no sign-in. The Free plan only needs you to accept the terms; paid plans are activated with a license key sent by email.

There is no public download yet. AI Translator is in beta: you request the macOS installer (macOS 14.2 or later, Apple Silicon) by email on the Download page. Windows 10/11 is coming soon, with no release date yet.

No. AI Translator is a commercial product used under the license in the Terms of use. It does use many open-source components from third parties, such as Whisper, whisper.cpp, llama.cpp, Hy-MT2 and Tauri. The full list of licenses is inside the app (About › Open-source licenses), with a summary on About AI Translator.

Five languages, for both the source audio and the translation: English, 中文 (Chinese), 日本語 (Japanese), 한국어 (Korean) and Tiếng Việt (Vietnamese). The app interface is in Vietnamese and English.

The app detects the spoken language among the ones you tick, or you can lock a single language. Translation quality has been measured for the eight directions that involve Vietnamese (English, Chinese, Japanese and Korean into Vietnamese, and Vietnamese into English, Chinese, Japanese and Korean). The other directions run, but we have not measured a quality score for them.

We plan to add more languages in the future. There is no schedule or list yet, and the app currently supports five languages: English, 中文, 日本語, 한국어 and Tiếng Việt. When a new language arrives, we will update this page and the Features page.

One direction: from the audio playing on your computer into your language. The app does not yet translate your voice to play into the meeting. If both sides install the app, each side sees the other's subtitles.

Not perfectly: translations can be wrong, incomplete or out of context, especially with technical terms, proper names and unclear speech. Do not rely on them for important decisions.

The glossary (a Pro feature) hints your names and terms to the translator, but does not guarantee it uses them correctly every time. We do not publish an overall accuracy rate because we have not measured it on real meeting conversations; the measurements we have, with their conditions, are on the Features page.

The Standard pack (about 2.5 GB) makes fewer recognition errors, especially in Vietnamese, Japanese, Korean and Chinese. The Lite pack (about 1.3 GB) is smaller and meant for machines with less RAM (8 GB minimum). The app recommends a pack based on your RAM and graphics card.

On a standard set of read-aloud clips, the word error rate for Vietnamese is 8.7 % with Standard and 22.5 % with Lite (Mac M4 Pro, read speech rather than real meetings). If you listen to these languages a lot and your machine has enough RAM (16 GB recommended), use Standard.

On a Mac M4 Pro, the median delay is under about 1.1 seconds from the moment a speaker finishes a sentence to the full translation appearing; the first translated words show after about 0.5 to 0.7 seconds, with either model pack. That is a measurement on one specific machine, using audio played back in real time.

We have not measured a base Mac M1, 8 GB machines or Windows discrete graphics cards. A preliminary test on a Windows laptop with integrated graphics showed that the Standard pack does not meet our delay target. Details and conditions are on the Features page.

No. Both speech recognition (Whisper) and translation (Hy-MT2) are AI models that run on your own computer. No cloud AI service takes part in recognizing or translating your meeting, so the audio, the transcript and the translation do not pass through any third party. The network is only used for things other than translating: checking your license, payments, downloading models and updating the app.

Entirely on your computer. Audio exists only in RAM while translating; the transcript and translation appear on your screen and live in the session's memory; history is off by default, and if you turn it on it is stored encrypted on your machine. Our server holds no conversation data, only your email (when you buy), orders, licenses and device-activation details (a hashed machine ID and device name). See Data and security.

Yes: speech recognition and translation run 100 % on your computer and need no internet once the models are downloaded. You still need the internet occasionally for a few things: the first model download (about 1.3 GB or 2.5 GB), the first Free trial registration, buying a plan, license checks (a paid plan works offline for at most 14 days between two checks) and app updates. See offline vs cloud translation.

No. Audio stays in RAM, is never written to disk and never sent over the network. We checked with a proxy and nettop on macOS: while translating, the app sends no audio or transcript text out. Windows has not been measured yet. Details are on Data and security.

There are no ads and no analytics, and the app sends no automatic crash reports. Our servers store only your email (when you buy), orders, licenses, a hashed machine ID, the device name (when you activate a key), the Free trial dates and a license change log, so we can issue keys and keep to one device per key and one trial per device. We do not sell your data or share it for advertising.

Not by default. The transcript of the running session is held in memory; you can view, search and copy it on every plan. Saving history (encrypted, on your computer only) and exporting to TXT, SRT or Markdown are Pro features. See the history and export guide.

We do not give legal advice. What we can say: AI Translator does not record meetings to disk; it only shows translated subtitles and, optionally, saves a transcript.

If the law or your company policy requires it, you are responsible for telling the other people in the meeting that you use a translation tool. The app reminds you at the Privacy step of first-run setup, and the Terms of use say so explicitly, together with the requirement to follow your organization's security rules and not to use the app to eavesdrop.

On your computer: Settings › Privacy › Delete all data (history and glossary), or Delete models and data. On our servers: email support@aitranslator.io.vn from the email you used to buy and ask for deletion or anonymization. We remove your email and device name, and still keep the hashed machine ID, the time of consent and the accounting-level order row. See how to request deletion.

The beta runs on Apple Silicon Macs (M1 or later) with macOS 14.2 or later; at least 8 GB of RAM, 16 GB recommended. Intel Macs are not supported. Windows 10/11 64-bit (CPU with AVX2) is coming soon; Windows ARM64 is not supported.

If a machine has less than 8 GB of RAM or does not meet the requirements, the app explains why and does not download models. For a base Mac M1 specifically, we have no measurements yet.

Models take about 2.5 GB of disk (Standard) or 1.3 GB (Lite), and you need 1 GB more free while downloading. On a Mac M4 Pro, the two engine processes use about 2.9 GiB of RAM (Standard) or 1.8 to 1.9 GiB (Lite), plus roughly 0.3 GB for the app. The engines shut down after 10 minutes without translating, so the app sitting in the menu bar does not hold several GB of RAM. We have not measured other machines.

Because the macOS build is currently ad-hoc signed and not notarized: we do not have an Apple Developer ID yet. macOS therefore blocks the first launch. Open System Settings › Privacy & Security, scroll to the bottom, click Open Anyway next to AI Translator and confirm with your password or Touch ID. From macOS 15, the right-click › Open trick no longer works. See the macOS installation guide.

Only System Audio Recording permission; AI Translator does not use the microphone. macOS asks the first time you press Start. If you decline, macOS shows no error and the app simply receives silence; the app will prompt you to turn it back on. See the audio permission guide.

On macOS, first choose Settings › Privacy › Delete models and data, then drag the app to the Trash (macOS gives no prompt when you remove an app). On Windows, the uninstaller is designed to include an option to delete app data, which also removes the models. Deleting data or uninstalling does not remove your license or the quota you have left.

Free is a trial, not a permanent free plan: 10 days from the first time a device registers for the trial, up to 30 minutes of translation per day, once per device. Uninstalling and reinstalling does not restart the 10 days. After the trial you need Monthly (50,000 ₫, 50 hours per 30 days) or Yearly (500,000 ₫, unlimited for 365 days). See pricing.

Only speech that has been translated is counted, by the length of the speech. Silence, sentences already in the language you want to read, and segments that could not be translated are not counted. The quota is per device, and the app warns you when less than 5 minutes remain.

Not yet. For now we accept only bank transfers from a Vietnamese bank, in VND, through VietQR (the PayOS gateway). The app draws the QR code right on the Upgrade screen; the code works for 15 minutes, and when PayOS confirms the payment the app activates the plan and your key is emailed to you. There are no recurring payments and no e-invoices yet.

No automatic renewal: each plan is a prepaid order that simply ends, and you are never charged again. The app reminds you 7 days ahead. At expiry the device goes back to Free if it is still within its 10-day trial; after the trial you need to buy a plan to keep translating, and Pro features (history, glossary, export) are locked. Renewing the same plan adds another 30 days (Monthly) or 365 days (Yearly).

Payments are not refundable, except when a fault on our side stops you from using the plan you bought and we cannot fix it within a reasonable time, or when the law says otherwise. Send the request within 7 days of the payment date to support@aitranslator.io.vn with the order code. Changing plans is not refundable. Details are in the Terms of use.

One key works on one device; reinstalling the app on the same device does not count as a new one. To move, on the old device go to Settings › License › Deactivate this computer; or on the new device enter the key and choose “Remove that computer and use this one”.

If two devices activate at once, the key is locked on both until one of them removes it. Changing devices too often within 30 days locks the key temporarily and you will need to contact support. The Free trial is per device, so it cannot be moved. See the guide to buying and activating a key.

Go to Settings › License › Lost your key?, enter the email you used to buy and choose Send my keys: every valid key for that email is sent to that same email. If that does not work, email support@aitranslator.io.vn from the email you used to buy.

You can adjust a lot. The bar always stays on top and does not take focus from your meeting app. Drag it to move it, drag an edge to resize it, and choose the font size (14 to 48 px), text color, background color and background opacity. Turn on Lock so the mouse passes through the bar; hide it quickly with the shortcut or the ✕ button when you hover over it. See the subtitle bar and shortcuts guide.

On macOS: ⌃⌥T starts or stops translating, ⌃⌥H shows or hides subtitles, ⌃⌥L locks or unlocks them, and ⌃⌥PageUp and ⌃⌥PageDown scroll to older and newer sentences (⌃ is Control, ⌥ is Option). On Windows it is Ctrl+Alt with T, H, L, PageUp and PageDown. You can change them under Settings › Shortcuts, so if one clashes with another app, just change it.

No. Closing the window only hides it to the menu bar (macOS) or the system tray (Windows); shortcuts and a running session keep going. To quit completely, choose Quit from the menu bar or tray icon. On a Mac, ⌘Q does not quit the app either.

Yes, through the Stable or Beta channel (Settings › General › Update channel). The app checks at launch and every 24 hours, verifies the signature before writing any file, and then invites you to restart when it is idle. An installer sent by email at the start may not auto-update yet. On the ad-hoc signed macOS build, each update makes macOS ask again for Keychain access and audio recording; the app warns you in advance.