Sazbauer

Every word you save can wear a picture. Choose where it comes from — found photographs, or a drawing made by an AI just for the word.

💬 Schwätzen is told roughly where you are, and aims there. It is not held to it by rule any more — the conversation was made stiff by being told how to talk, so it is now only told about you.

Only you can judge this one. Correcting every slip teaches the most and interrupts the most; correcting what matters keeps the conversation moving. Change it whenever it starts to grate.

When you talk hands-free, silence is how the app knows your turn is over. Too short and it cuts you off while you are still hunting for the word; too long and it feels slow.

By voice, the chat is already listening when you open it — you just start talking. By tap, the microphone stays off until you press 🎙.

When you speak to the app — the conversation, the spoken answers in the game, the pronunciation check — this recogniser listens. A short live clip is its own job: the ear for imports is chosen separately below.

When something is imported — a pasted link, an upload, a podcast — this recogniser transcribes it. LuxASR is strong here, and it is the one that keeps speaker turns. The microphone above is chosen on its own, and the voice that speaks below.

The mouth, chosen on its own: the voice that reads replies, corrections and words aloud. The ear above listens with whichever recogniser you picked — the two no longer travel together.

The model behind 💬 the talk, 🔁 the interpreter and a typed 🤖 line — it writes the reply and, when corrections are on, marks your sentence in the same breath. Everything else in the app that asks the AI in passing keeps its own model; this is the conversation only.

A different kind of model entirely: it hears your voice and answers in its own, with nothing written down in between. Used only by 🎙 in 🤖 mode. Changing it takes effect on the next call — and remember a call is billed by the minute of audio, not by the turn.

Dresses both halves of 🤖: the live call, and the voice that reads a typed reply aloud. Luxembourgish is never spoken by these — that stays D'Sproochmaschinn's, above.

A creator's clips arrive by the hundred, and every stored one takes space that is never backed up. By default nothing is kept: the video is fetched from TikTok each time you play it. The words, the sentences and their timings are always yours either way — and the lyrics page, tap sync and the karaoke captions work the same on a borrowed clip as on a stored one. Podcasts, uploads and other links are never touched.

Everything — texts, sentences, sets, words, progress — as one file on this device. Your account, in your own hands.