Qenzo AI dictation keyboard: speak one language, send another

How the switch actually works

Two settings apply to every dictation, and they are independent of each other. The first is the language you spoke — either named directly, or left on auto-detect so Qenzo works out which language it was without being told. The second is the language the finished message comes out in, chosen separately for that one message. Neither setting is a keyboard you switch or a mode you turn on for a session: you can dictate a message into Russian one minute and English the next, from the same keyboard, without changing a setting that carries over to the next thing you type.

What happens in between is not translation of a transcript. Qenzo does not write down what you said and then run that text through a translator; it goes from understanding the speech directly to finished text in the language you picked, so the result reads the way a person writing that language would write it, not the way a machine-translated sentence reads. An "as spoken" option turns this off entirely and keeps the output in the language you spoke, for the messages where that is what you want instead.

Mixed-language speech survives the same way. Half a sentence in one language and half in another — "so yeah I called them and asked, но они сказали что перезвонят завтра" — is normal speech, not an error, and it still comes out as one clean message in whichever single language was chosen: the language switch itself disappears, the false start disappears, and the fact underneath it (they said they would call back tomorrow) stays. The recording is also checked a second time against the finished text, so a name or a number gets corrected instead of guessed, and the original transcript can be opened to see what was actually said. That second check is skipped only when a tone, recipient or role rewrite is layered on afterward, since a rewrite replaces the text rather than re-checking it against the audio.

Seven pairs, worked — and two of them the other way round

The mechanism does not change based on which two languages are involved. These are ordinary, low-stakes messages, the kind that make up most dictation, shown as what was actually said and what Qenzo wrote.

• Russian to English. Spoken: "Извините, я опаздываю на созвон минут на десять, буду скоро." Written: "Sorry, I'm running about ten minutes late for the call — I'll be there soon."
• Armenian to English. Spoken: "Ներողություն, մի փոքր ուշանում եմ, մոտ տասը րոպեով։" Written: "Sorry, I'm running a little late — about ten minutes."
• Armenian to Russian. Spoken: "Շնորհակալ եմ օգնության համար, ինձ շատ օգնեց։" Written: "Спасибо за помощь, это мне очень помогло."
• Spanish to English. Spoken: "Oye, te mando el documento en un rato, dame diez minutos." Written: "Hey, I'll send you the document shortly — give me ten minutes."
• Indonesian to English. Spoken: "Maaf, saya masih di jalan, mungkin telat sepuluh menit." Written: "Sorry, I'm still on my way — might be about ten minutes late."
• Arabic to English. Spoken: "آسف، سأتأخر حوالي عشر دقائق، أنا في الطريق." Written: "Sorry, I'll be about ten minutes late — I'm on my way."
• Hindi to English. Spoken: "माफ़ करना, मुझे पहुँचने में दस मिनट और लगेंगे।" Written: "Sorry, it'll take me about ten more minutes to get there."
• English to Russian (the other direction). Spoken: "Thanks for waiting — I'll call you back in five minutes." Written: "Спасибо, что подождал, я перезвоню через пять минут."
• English to Armenian (the other direction). Spoken: "Can you send me the file whenever you get a chance?" Written: "Կարո՞ղ ես ուղարկել ֆայլը, երբ հնարավորություն ունենաս։"

All 42 languages Qenzo can send a message in

The same 42 languages work on both sides: what you can speak into the microphone and what a message can be written in are the same list, chosen independently for each message. The list below holds 43 entries rather than 42, because simplified and traditional Chinese are two ways of writing one language and count once between them, while Cantonese is a different spoken language and counts on its own.

• Arabic
• Armenian
• Bengali
• Cantonese
• Chinese (Simplified)
• Chinese (Traditional)
• Czech
• Danish
• Dutch
• English
• Finnish
• French
• German
• Greek
• Gujarati
• Hebrew
• Hindi
• Hungarian
• Indonesian
• Italian
• Japanese
• Kannada
• Korean
• Malay
• Malayalam
• Marathi
• Norwegian
• Persian
• Polish
• Portuguese
• Punjabi
• Romanian
• Russian
• Spanish
• Swedish
• Tagalog
• Tamil
• Telugu
• Thai
• Turkish
• Ukrainian
• Urdu
• Vietnamese

What this is not

Qenzo is not a translator app that you paste text into. There is no separate app to open and no text to copy: it replaces the phone's keyboard (and, on the desktop client, works from a key you hold down instead), so speaking in one language and getting text in another happens inside whatever you were already typing into — WhatsApp, Gmail, Slack, a note, anything with a text field.

It is also not a live interpreter for a conversation. Qenzo turns one person's dictation into one finished, written message; it does not listen continuously to two people talking and speak a translation back in real time, and it is not built for a back-and-forth exchange. Nor is it a file transcription service — each dictation is something spoken live into the microphone by the person using Qenzo, not an audio file uploaded afterward for a transcript. Each recording is understood once and written once, in the language chosen for that message.