Hold one key to dictate, two keys to translate
Half your day is in one language, half in another. Slack to colleagues in English, voice notes to your mother in Spanish, an email to the Berlin client in German. You think the sentence in one language and then retype it more slowly in the other — the slow tax that makes bilingual work feel not quite worth the speed of voice.
You hold the same key you have always held to dictate. You press a second key while you talk. Lispr types the translation at your cursor, in the app you were already in.
The two-key model
Lispr has one trigger key for dictation. By default it is the Option key. You hold it down, you talk, you release. The text appears where the cursor was.
Translate is a second key, held in addition to the first. By default the Control key is one target language, the Shift key is the other. You hold Option to start dictating, you press Control mid-sentence, you keep talking, you release. What you said in Spanish comes out in English. Your finger gesture is one extra modifier — a chord, not a menu.
The second key does not change which language you are speaking. Lispr detects that automatically. The second key picks the language you want Lispr to type in.
The two slots
There are exactly two translate slots. Each slot is one modifier key bound to one target language. You set them once, in Settings, and forget them.
Out of the box Lispr ships sensible defaults based on your system language — English-locale users get Spanish and French, Spanish-locale users get English and Portuguese — but the two slots are yours to repoint. A reasonable pair for the bilingual day above:
- Control → English (the language of work)
- Shift → Spanish (the language of family)
Now their two free fingers each carry a language. They hold Option and talk. If they press nothing extra, they get the language they spoke in, typed verbatim. If they press Control, they get English. If they press Shift, they get Spanish. There are thirty-two target languages to choose from, including the major European, Slavic, East Asian and Middle Eastern languages.
How to set them
Click the Lispr icon in the menu bar and choose Translate from the dropdown. A small window opens with two boxes — one per slot. Each box has two dropdowns: which modifier key holds the slot, and which language that slot translates to. Pick a key, pick a language, close the window. Done.
Defaults are already filled in the first time you open the window, based on your system language. There is no onboarding step. If they fit your day you can leave them and move on; if not, swap them in twenty seconds and forget about them.
Pressing the key mid-sentence
The translate key does not have to be down from the start. You can hold Option, begin a sentence in Spanish, realize halfway through that this one is going to the English channel, press Control without releasing Option, finish the sentence, and release Option. Lispr translates the whole utterance — not just the part after you pressed Control.
The reverse also works. You can hold Option and Control together from the moment you start talking, and the result is the same. Both styles end up at the same place. The key is a verdict, not a switch.
There is a small grace window. If you let go of the translate key a beat before the trigger — which is what fingers actually do when they release a chord — Lispr still counts it as a translate. You do not have to be neat about it.
If you press the slot key and then change your mind, release it before you release the trigger. The recording goes through as plain dictation. Only the translated text is typed at your cursor — never both. Standard Cmd-Z undoes it like any other typed text.
Where you see what is happening
While you are holding both keys, the Lispr pill on screen shows a small badge with the target language code next to the waveform. →EN if you are translating to English, →ES for Spanish, →UK for Ukrainian. The badge stays through the brief transcribing moment after you release. It is a quiet confirmation that what is about to land at your cursor is the translated version, not the verbatim one.
How fast it actually is
Translation runs on a server. End to end, from releasing the key to seeing the text, you are looking at about half a second for a short message, up to seven hundred milliseconds for a paragraph. By the time your hand has settled and you have looked back at the cursor, the text is there — for subsequent translates in a session. The very first translate after waking the app may take a touch longer while the connection warms up.
What it is good for
This is a tool for short and medium-length working language.
- Slack and chat replies. "On my way home, twenty minutes" said in English, typed as Spanish for your mother: "Voy camino a casa, veinte minutos." The other direction works the same — you say "Voy de camino — llego en cinco minutos" in Spanish, Lispr types "On my way — five minutes out" into the English Slack channel.
- Email in a language you read fluently but write slowly in. You speak the English email and press the German slot. You go from a blank box to a paragraph in the time it takes to think the sentence through once.
- Replying to voice messages in WhatsApp or Telegram. Listen to the Italian voice note, hold the trigger and the Italian slot, speak your answer in English, release. The Italian answer is typed into the reply field.
- Drafting in your native tongue for a localized document. You write the Ukrainian version of a doc by speaking Ukrainian and dropping into the Polish doc with the Polish slot. Each paragraph goes from your head to the page in one language pass.
The translations are good in a way that matters for these uses. They preserve tone — formal stays formal, casual stays casual. They keep numbers and dates exact. They keep proper nouns in place, including names you have taught Lispr in your custom vocabulary. In our internal blind A/B benchmark across dozens of mixed sentences, the engine scored 4.76 out of 5 with zero catastrophic mistranslations.
What it is not for
- Not a translator for contracts or anything legal. A fast LLM is good at register and meaning; it is not a sworn translator. Read the output if the stakes are real.
- Not a poetry engine. The model carries meaning across cleanly; it does not carry rhyme, meter or the particular weight of a word. For a literary translation, write it yourself.
- Not a language-learning tool. You only see the output. You do not see what you said. If you want to learn the target language, dictate verbatim in that language and read what comes back.
- Not a way to translate text you already have on screen. Lispr only translates what you say. There is no camera, no OCR, no "translate this selection." You speak the source — even if you are reading it off a page — and Lispr types the target.
- Not offline. If you are on a plane with no Wi-Fi, dictation doesn't work either — Lispr is cloud-based end to end.
If the translation call ever fails — bad network, server hiccup — Lispr does not throw away your words. It types the original transcript at your cursor instead. You may have to translate that one by hand, but you do not lose what you said.
Try it for a day
Open the menu bar, pick Translate, and set the two slots to the two languages you actually need most — your working language and your home language, most likely. Use it for a day. If a phrase comes out wrong in a way that makes the message land badly, drop us a note through the Contact link in the site footer and paste both versions.
Press your Lispr key, press the second one for the language you need, and trust the next half second to carry your sentence across.
Try Lispr
Voice to text in any Mac app — hold a key, talk, let go. Free, no account, ~4 MB.
Download for macOS