Frequently asked questions

Three steps: download and install the Windows desktop app or the Android app from the download page, sign up and finish binding your account inside the app, then you can start talking. Without a bound account, the app blocks the hotkey and won't record, so this step can't be skipped. New members get a 7-day full trial, no credit card and no payment needed up front.

By default, press the right Alt key: press once to start recording, press again to stop and send it for recognition. To use a different key, open the settings window and click "Change hotkey…", then press the key you want. While recording, a red status overlay on screen counts down the seconds remaining; a single recording can run up to 120 seconds (2 minutes).

A floating circular button, 60dp across, sits on top of every app. You can drag it to wherever suits you and it remembers where you left it, and it automatically moves out of the way when the keyboard pops up. After 3 seconds idle it fades to 35% opacity so it doesn't block the screen — tap it once and it comes back. Tap it to start talking; while recording the whole button inverts to a black background with a white logo and a progress arc traces around the edge, and while it's processing it fades in and out gently.

Yes. 聆打 Linda isn't tied to any one app — LINE chats, notes, browser forms, expense apps, anywhere there's a text field. The text is typed straight into wherever your cursor is: on Windows it's pasted from the clipboard (falling back to typing it out character by character if the clipboard isn't available), and on Android it's written in through the accessibility service.

On Android, if no text field is detected on the current screen, the recognized text is copied to the clipboard instead, and you just long-press to paste it wherever you need — so nothing you said is wasted. That means it still works in games, image viewers, or apps with custom-drawn interfaces; you just say it first, then paste it where you want.

With cloud recognition the limit is 120 seconds (2 minutes) per recording, the same on Windows and on Android in cloud mode. Once you turn on Android's "on-device recognition", that limit stretches to 600 seconds (10 minutes), which is handy for getting a long passage out in one go. The time remaining shows as a countdown on the Windows overlay, and as a progress arc around the Android button.

With cloud recognition it's usually a matter of seconds, though the exact timing varies with your network and server load, so we'd rather not promise a specific number we can't guarantee. With Android on-device recognition, it streams as you speak, so the text appears almost the instant you stop talking, because the whole process happens on the phone without going over the network.

Right now there are two versions: the Windows desktop app (needs Windows 10/11 64-bit) and the Android app (needs Android 8.0 or later). There is no macOS, iOS or Linux version, and no browser extension. Install steps and system requirements are listed on the download page.

Not yet, and we'd rather not set the wrong expectation about something that doesn't exist. If you have a Windows PC or an Android phone, you can start with either of those — the account is shared, so binding on either platform counts as the same account.

The Windows desktop app is currently on version 1.0. Once it's installed, you don't need to keep track of version numbers yourself — it checks for updates at most once a day and downloads any new version so it's ready the next time you start it.

The Android app is currently on version 1.1 (versionCode 101). Version 1.1 brought a visual refresh (cream color scheme, pill-shaped buttons, group cards), and the personal dictionary now works on a 500-character total budget, showing how much you've used so far.

One account can bind up to 5 devices, mixing computers and phones however you like, with up to 3 recognitions running at the same time. Reinstalling the Android app doesn't use up an extra seat — the system recognizes it as the same phone. Accounts are for personal use, so please don't share one across multiple people.

Click "Unbind" on the old device to free up the seat, or unbind it from the devices page in the member center on the website. After unbinding, you need to wait 24 hours (1 day) before rebinding, and each account can rebind at most 3 times a month. If your phone is lost, you don't need to change your password — just unbind it from the member center.

It needs Android 12 (API 31) or later, and your phone needs to have the offline speech pack for that language already installed. If your phone doesn't meet those conditions, the toggle is locked off and shows "This device doesn't have the offline Traditional Chinese model, so it can only use the cloud"; while the model is downloading you'll be told to stay connected, since it still uses the cloud until the download finishes. The model code for Traditional Chinese is cmn-Hant-TW.

By default no transcripts are kept — voice is used only to do one thing: turn it into text. On Android, if you turn on "on-device recognition" and turn off "polishing mode", neither the voice nor the text ever leaves your phone; if on-device recognition is on but polishing is also on, only the text is sent to the cloud for tidying up. Full details are in the Privacy policy.

No, and we want to be clear about this: speech recognition on the Windows app is actually sent to a cloud server, and there is currently no working offline mode for Windows. An old version of the app's settings window has a line that says "speech is always recognized locally", which doesn't match what actually happens — that's a copy error on our part and we're correcting it. If you need voice to never leave the device, use Android's on-device recognition instead.

On Android, yes: turning on "on-device recognition" means audio isn't uploaded, doesn't use mobile data, and even works in airplane mode. It requires Android 12 or later with the offline speech pack for that language already installed, and offline mode doesn't support Taiwanese. The Windows app currently has no offline mode at all — it always needs a network connection to recognize speech.

When both "on-device recognition" is on and "polishing mode" is off at the same time, yes, nothing leaves the device. The privacy status line in settings reflects this combination in real time, showing "Fully offline: voice and text never leave your phone" when that's the case. If it says "Mixed" or "Cloud" instead, something is being uploaded, and you can always check which state you're in.

No. Your personal dictionary is tied to your own account and only syncs between your own devices — Windows, Android and the website member center all support automatic sync, and opening the dictionary page on Android triggers a sync too. It's not a shared public list; other people's words never end up in your dictionary.

No. Polishing only changes "how it's said", not "what was said" — it adds a missing subject, reorders words, and trims off the bits where you changed your mind mid-sentence, but it never adds content you didn't actually say. If you don't want any tidying up at all, turn polishing off and use verbatim mode, which only strips out filler words like "um" and "uh".

The commission rate is 30%. Every time someone you referred is charged their NT$300 monthly fee, you earn NT$90, and it's a recurring commission, not a one-off. To be clear, it's not open-ended — each referred member generates commission for you for up to 6 months from their first payment. You also need to be a paying monthly member yourself for commission to start counting.

The math is straightforward: divide your own monthly fee of NT$300 by the NT$90 each referred member brings in per month, and that's how many you need. There are two conditions — you need to be a paying member yourself, and your referrals need to keep paying too — and each person's commission only counts for up to 6 months, after which you need new referrals to keep it up.

Commission has a 30-day lock-in period after it's generated, and it automatically becomes withdrawable once that's over. The minimum payout amount is NT$2,000; once you submit a withdrawal request, it's transferred by bank transfer manually within 3 business days, with a NT$15 transfer fee paid by the recipient. Commission counts as Taiwan-sourced income, and a withholding tax statement is issued once it reaches the withholding threshold.

Referrers in Taiwan withdraw by bank transfer only, and the account name has to match your own — it can't be paid to someone else. PayPal is only available to overseas dealers, because Taiwan regulations have not allowed PayPal payments within Taiwan since 2017 — it's not that we don't want to offer it.

A sign-up counts as yours if it happens within 90 days of someone clicking your referral link. Attribution is first-touch — meaning a person is permanently tied to whichever referrer's link they clicked first, and clicking someone else's link later doesn't change that. So getting your link out early does matter.

No, self-referral is blocked — signing up a second account with your own link doesn't generate commission. Commission is designed to reward people who genuinely bring the product to others, not as a way to discount your own account. If you're looking to save money, buying a longer plan on the Pricing page is the more direct way.

Yes. Your commission wallet can be used two ways: withdrawn as cash, or applied directly to your own monthly fee. If you haven't reached the NT$2,000 payout threshold yet, applying it to your monthly fee is the most direct way to use it without waiting.

Yes, the system supports per-channel referral codes — one for a Facebook group, one for business cards, one for a LINE group, each tracked separately. That way you can see which channel is actually bringing people in and focus your effort there. Links can be shared even before you start paying yourself — clicks and sign-ups are recorded either way.

A few things that are fairly uncommon: 12 recognition languages including Taiwanese and Cantonese, longer recording limits than most voice keyboards (120 seconds in the cloud, 600 seconds on Android on-device), corrections you make in the preview window get learned into your personal dictionary automatically and synced to your other devices, and on Android you can get voice and text to never leave the device at all. We're a small, early-stage team, so we're not going to claim we're "industry leading" — the best way to compare is to try the 7-day trial yourself.

A voice input method needs you to switch to it before you can use it; 聆打 Linda is a floating button or a hotkey that types straight into whatever field you're already in, so your existing input method never has to change. It also comes with things a typical voice keyboard doesn't have, like a personal dictionary, a verbatim/polished toggle, and multi-language selection.

A single month is the standard price of NT$300; buying three months, six months or a year at once all come with a discount, and the longer the period, the lower the average monthly cost. The annual plan saves the most, but if you're still deciding whether it's the right fit, trying the 7-day trial first and then buying a single month to observe is a lower-pressure way to start. See Pricing for the exact amounts for each plan.

If your phone is on Android 12 or later and has the offline speech pack installed, on-device recognition is the better default: a 600-second limit, text appears almost the instant you stop talking, and with polishing off, nothing leaves the phone at all. On the other hand, if you need Taiwanese, your phone doesn't meet the requirements, or you're on the Windows app, you're limited to cloud recognition with its 120-second limit.

It depends on whether you'll edit it yourself afterward. If you want to keep the exact wording to organize later (interview notes, class notes), use verbatim; if you want something ready to send as-is (replying to a client, writing a daily report), use polished. If you're not sure, start with verbatim, and switch on polishing if you find there are too many loose ends — you can switch between the two any time.