Lingua turns your iPhone into a hands-free simultaneous interpreter, a personal AI memory and a discreet voice assistant — built on realtime speech-to-speech neural networks.
Everything streams live over secure WebSockets to realtime AI models — no push-to-talk, no waiting for whole sentences.
Two people, two languages, one phone. You speak — the translation plays out loud on the speaker for your interlocutor. They reply — you hear the translation privately in your Bluetooth headphones. Both sides see live subtitles. Automatic language detection chooses the direction; hardware echo cancellation keeps the loudspeaker out of the microphone.
realtime speech-to-speechA discreet "invisible earpiece" mode: Lingua listens to the conversation in the room and, when a question is addressed to you, instantly whispers a short, ready-to-repeat answer into your earphone — phrased in first person, in the right language. It never interrupts and stays silent on small talk.
whispers in ~2 secondsDictate anything — "dentist tomorrow at 10", a shopping list, a dream, a document number. Live on-device captions show what you said; an AI brain files it into the right category, creates reminders (synced to the native iOS Reminders app) and makes it searchable forever.
voice notes → organized knowledgePoint the camera at a menu, sign or contract — or import photos and PDFs from your gallery and Files. Lingua OCRs the document, translates it into your language and stores the full text in your memory for later search.
photo · gallery · PDFA natural speech-to-speech assistant that actually knows you: it answers from your private memory — notes, reminders, scanned documents and connected email — using retrieval-augmented search, and reaches the live internet for fresh facts, news, prices and weather. Ask it to remember something and it files a note or sets a reminder by itself.
private RAG + live web searchConnect Gmail and your inbox becomes part of your memory: new mail is fetched automatically, indexed semantically, and the assistant can quote it back when you ask "when is my flight?" or "what did the landlord write?".
Gmail sync & semantic searchThe whole pipeline is streaming — audio leaves the phone as raw PCM frames and translated speech starts playing back while the sentence is still being analysed.
The iPhone records 24 kHz audio with hardware voice processing (echo cancellation), so the loudspeaker never feeds back into the mic.
Audio streams over an encrypted WebSocket to the Lingua backend (EU cloud), which relays it to realtime neural speech models.
A speech-recognition model transcribes and detects the language on the fly; the engine picks the translation direction per phrase — no buttons, no toggles.
A speech-to-speech translation model generates natural spoken output, routed to the loudspeaker or your Bluetooth earpiece depending on who needs to hear it.
Pick your language and your interlocutor's language once; everything else is automatic.
Cross-script pairs (e.g. English ↔ Russian, English ↔ Chinese) are detected instantly by script analysis; same-script pairs (e.g. Spanish ↔ English) use a statistical language identifier — committed once per phrase so the audio never flips mid-sentence.
Lingua is built on Microsoft Azure AI infrastructure in the EU, combining several specialised models into one streaming pipeline.
Azure OpenAI realtime models handle live voice translation, the voice assistant and the AI prompter: audio in, natural translated/spoken audio out, with server-side voice activity detection and tool calling.
A realtime Whisper model produces live captions of the source speech and powers automatic language detection that steers the translation direction.
Categorises dictated notes, extracts dates and reminders, parses and translates scanned documents (OCR), and structures email content — with controllable reasoning effort for speed.
Notes, documents and emails are vectorised with text-embedding models into a per-user semantic index; the assistant retrieves only your own data, scoped by your account.
For fresh public facts — news, prices, weather, schedules — the assistant queries a Google-backed search API and answers with up-to-date information.
Swift + AVAudioEngine with hardware echo cancellation, per-direction audio routing (loudspeaker vs Bluetooth), on-device live captions via Apple Speech, and EventKit integration for native reminders.