Instead of typing: the mic button in the AI
German speech recognition
Typing in the AI window is a different ballgame than speaking. I can write a correct sentence but not say it out loud. That’s why there’s a big button in the AI: hold, speak, release.
The Gesture
I press, speak, let go. If the sentence is going in the wrong direction, I pull my finger away. The attempt is gone. No embarrassing half-sentence that the AI takes seriously. That’s the whole trick. No menu, no setting I have to search for every time.
The recognition isn’t magic. It swallows endings. I say the sentence again, shorter. The meaning is enough. Anyone waiting for a perfect transcript won’t get past the second sentence.
Before the First Person
The button is my warm-up. Role, photo, brief scenario. It’s free, with no token display in the app. If I have my own key, it stays on the device. It’s not required.
Five minutes are enough. After that, I know if my mouth is cooperating today. If not, I’ll stick to the cards. If so, the voice group isn’t a leap into the unknown.
What I Don’t Do
I don’t dictate essays. One sentence, one answer, another sentence. The AI is supposed to ask questions, not give a lecture that I just nod along to. That’s what the four modes are for: Role, Photo, Debate, Situation. The button is the same in all of them.
Errors Out Loud, Not in the Text Field
A typo can be deleted without me even hearing it. A spoken error lingers in my ear for a second. That second is the lesson. I correct it once and move on. Three times the same syllable—then it’s stubbornness.
Afterward
The group gets the same sentence the button has already heard. Not a new, bold one that I’ve never practiced.
An Evening with the Button, Specifically
I open the AI, not the chat with people. Role mode, small scene: baker, one loaf of bread, one price. Taste, “one loaf of bread, please,” let go. If the ending is missing, try again, shorter. Then the follow-up question. Five exchanges, not fifteen. A photo only if the role doesn’t work today: a picture of the table, two sentences describing what’s on it. I put off the debate until the “because” is no longer stuck in my throat.
The recognition sometimes spits out nonsense. I don’t argue with the line. I repeat the sentence exactly as I mean it. The point is to speak, not to keep a record. Whoever corrects the transcript while keeping their mouth shut is back to typing.
I don’t see tokens in the app because no packages are sold here. If the response slows down, it’s a technical glitch—no reason to buy credit. My own key, if I have one, stays in the device. I don’t post it in a group or in a note that gets synced.
Next, the three cards that appeared in the deck. SRS will bring them back in three days. The button heard the sentence today. The card ensures that I don’t have to wait a whole month to revisit it.