Voice typing has a quiet problem. The words you say are the same, but the place you say them changes what good text looks like. "hey can you check the build when you get a sec" is fine in a team chat. Pasted into a client email, it reads as careless. Pasted into a code comment, it is noise.
Context-aware dictation (also called app-aware formatting) fixes this by cleaning your speech differently depending on which app has your cursor. This guide explains what that means, how it works, what to check before you trust it, and how to set it up so voice typing in any app on Windows, macOS or Linux gives you text you don't have to rewrite.
What is context-aware dictation?
Plain definition: context-aware dictation is speech-to-text that looks at the active app or window and formats the transcript to suit it. The transcription step turns your voice into words. A second step, usually an AI language model, removes filler words, fixes punctuation and adjusts the style for the destination.
Basic dictation, like the built-in Win+H tool on Windows 11, gives you one style everywhere. That is often fine for short notes. The difference shows up when you move between email, chat, documents and code many times a day.
The same sentence, four destinations
Say this out loud: "um so the export bug is fixed in the latest build can you uh test it on your side before friday"
- Chat (Slack, Discord, WhatsApp, Teams): short and casual. "The export bug is fixed in the latest build. Can you test it on your side before Friday?"
- Email (Outlook, Gmail, Thunderbird): professional and warm, full sentences, no slang. A good email profile should not invent a subject line or a greeting you never said.
- Document (Word, Google Docs, Notion, Obsidian): formal, well-structured prose that reads like a status note.
- Code (VS Code, IntelliJ, terminals): comment style, and spoken identifiers such as "export handler" can become exportHandler or export_handler.
None of these needs a different sentence from you. The cleanup step does the adjusting.
How it works, step by step
- You press a shortcut and speak. The app records until you press it again or stop talking for a few seconds.
- The speech engine transcribes. This can be a hosted model, your own API key with a provider such as Groq or OpenAI, or an offline model running on your computer.
- The tool detects the active app. It reads which window has focus (for example Outlook or VS Code) and picks a matching profile.
- An AI model formats the text. It removes "um" and "uh", adds punctuation and applies the profile's style. It should keep your language: if you spoke Spanish, the text stays Spanish.
- The text is typed at your cursor. No copying and pasting between windows.
What to check before you rely on it
App-aware formatting is useful, but only if it is predictable. Before you pick a tool, check these points.
- Does it ever add things you didn't say? A formatter that invents greetings, sign-offs, hashtags or emojis creates more editing work, not less.
- Does it keep your language? Some AI cleanup quietly translates mixed-language speech into English. Look for a tool that keeps the language and script you spoke.
- What happens if the AI is unreachable? The safe behavior is to paste the raw transcript with a warning, so you never lose what you said.
- Can you write your own instructions? A custom prompt such as "Rewrite concisely and professionally" helps when the built-in profiles don't match your work.
- Can you choose the model? If you use a local model through Ollama or LM Studio, or your own API key, you control where the text goes.
- Do names and jargon survive? Formatting can "correct" a product or person's name into a common word. A custom vocabulary with "heard as / write as" corrections that runs before and after formatting protects those spellings.
Common profiles and what each should do
- Email: professional, warm, complete sentences; no invented subject or greeting.
- Chat: casual, short, light punctuation.
- Code: comment or docstring style; spoken identifiers turned into camelCase or snake_case.
- Document: formal, structured paragraphs.
- Spreadsheet (Excel, Google Sheets): cell-friendly text; numbers as plain numbers, short labels.
- Search (Google, Bing, DuckDuckGo): filler removal only, so the query stays a query.
- Social (X, LinkedIn, Reddit): punchy and natural, with no hashtags or emojis added.
- Default: light cleanup for every other app.
Setting it up in practice
You can build this workflow with a dictation app that has AI formatting per app. InsertKit, a voice typing app that also includes a text expander and clipboard history, is one example. Its Smart Formatting setting detects the active app and applies profiles like the ones above on Windows, macOS and Linux:
- Install the desktop app and sign in. The InsertKit Free engine needs no API key and has a daily limit.
- Turn on Voice Typing and keep the default shortcut (Right Alt) or choose another.
- Switch on Smart Formatting. Leave the automatic per-app profiles on, or add your own instructions.
- Add brand names and jargon in the Vocabulary tab, with corrections like "insert kit" → "InsertKit".
- Test the same sentence in your chat app, your email client and your editor, and compare the results.
On Android the idea is the same, but you pick the formatting style yourself (Clean-up, Email, Chat, Document, Social Post, Code or Search Query) instead of the app being detected automatically.
FAQ
Is context-aware dictation the same as translation?
No. It changes style and punctuation, not language. A well-behaved formatter keeps the language you spoke, and voice typing tools of this kind often support 100+ languages.
Does it work offline?
Transcription can run offline with a local Whisper model on desktop. The formatting step needs an AI model, which can be a local one through Ollama or LM Studio, or a cloud model. If no model is reachable, the raw transcript should still be pasted.
Will it rewrite what I meant?
It shouldn't change meaning. If a profile adds or drops content, use your own instructions or a lighter profile for that app.
Do I need it for short notes?
Not always. Built-in dictation is enough for quick notes in one app. App-aware formatting pays off when you dictate across email, chat, documents and code in the same day.
Wrap-up
Good dictation is not only about hearing the right words. It is about writing them the way the destination expects. Check that a tool keeps your language, never invents content and falls back to the raw text, then let it handle the style switch between apps. If you want to try the setup described above, you can download InsertKit for Windows, macOS or Linux.
Disclosure: this post was written by the InsertKit team. The checklist above applies to any dictation tool that offers per-app formatting.
Top comments (0)