Sprecho does not paste a transcript into your document. It pastes the text you would have written — punctuated, capitalised, broken into paragraphs, with the "ähm"s taken out.
This article walks through the four steps that happen while you are still letting go of the key, and draws a clear line between what the AI is allowed to change and what it must leave alone.
What happens between speaking and the text appearing
- You speak. While you hold the dictation key, Sprecho streams your audio to its servers in Germany.
- Speech recognition writes it down. The result is a raw transcript: your words, in order, with no punctuation and every filler intact.
- A formatting model turns spoken form into written form. This is the step that adds the commas, the capital letters and the paragraph breaks.
- The finished text is typed into your app — wherever your cursor was when you started.
Steps 2 and 3 both run on Sprecho's servers, which is why dictation needs to be online. Which model runs in step 3 depends on the app you are dictating into; see Formatting models.
What formatting actually changes
The formatting pass is a spoken-to-written conversion. Concretely, it:
| Changes | Example |
|---|---|
| Punctuation | "wir treffen uns morgen passt dir das" → "Wir treffen uns morgen. Passt dir das?" |
| Capitalisation | Sentence starts, and German nouns |
| Fillers | "ähm", "äh", "um", "uh", stammers, doubled words, abandoned sentence starts |
| Paragraphs | A blank line at topic shifts, roughly every two to five sentences |
| Spoken punctuation | "Komma", "neuer Absatz", "Klammer auf" become ,, a blank line, ( |
| Numbers and dates | "zwanzig Euro" → "20 €", "neun Uhr dreißig" → "9:30 Uhr" |
| Lists | Three or more items after a lead-in become bullet points |
| Spelled-out letters | "b g b" → "BGB" |
| Self-corrections | "Montag, nein, ich meine Dienstag" → "Dienstag" |
Greetings and closings get their own lines, and file names, paths and commands you speak are wrapped in backticks so they survive a paste into a chat window.
What formatting never changes
This is the part that matters most, because a dictation tool that quietly "improves" your text is worse than useless in a medical report or a contract.
- Your meaning. The model edits form, never content. No fact, condition or nuance is dropped, and nothing is invented.
- Names. Copied letter-for-letter. Never re-spelled, shortened or translated.
- Numbers, and technical, medical and legal terms. Kept exactly as spoken.
- Your grammatical person. "du" stays "du", "Sie" stays "Sie". An instruction to someone stays an instruction.
- Your order. Sentences and side remarks appear in the order you said them.
- Your voice. Profanity, emoji, hedges and tone particles stay in.
The model treats your dictation as text to clean, never as instructions to follow. If you dictate "write my colleague a short summary", you get that sentence back, punctuated — not a summary. Sprecho is a dictation tool, not a chatbot.
Where the raw transcript goes
Every dictation lands in the history on the Home screen with two versions: Formatted and Raw. The raw version is the untouched recognition output.
Use it to check what actually reached the recogniser when a word came out wrong — that usually tells you whether the problem is the microphone, the recognition, or the formatting. See Notes and dictation history.
When formatting gets a word wrong
If the same name or technical term comes out wrong every time, formatting is not the place to fix it — the recogniser never heard it correctly in the first place. Add the word to your dictionary and it is fed to the recogniser before your next dictation.
If the text is right but the tone is wrong for the app you are in, switch the model for that app category instead. Sprecho already knows whether you are in a messenger, an email client or a code editor — see Set a writing style per app.