Voice to Text for AI Prompting on Windows

By /

If you work with AI assistants every day, you know the pattern: the better your prompt, the better the answer. But a good prompt is long. It needs context, constraints, the error message, what you already tried and what the result should look like. Typing all of that takes time, so prompts get shorter and answers get worse. Voice to text for AI prompting fixes exactly that: you say what you mean, the text lands in the chat or editor, and you send a complete prompt in a fraction of the time. This guide shows you how to set it up on Windows, step by step.

Why speaking your prompts works so well

When you explain a problem to a colleague, you naturally give context. When you type the same thing, it often shrinks to „form saves twice, fix“.

Speaking lowers the cost of detail. A few extra sentences cost you a few seconds, not a minute of typing. And because AI assistants handle loose, conversational language well, your prompt does not need to be polished. It needs to be complete.

There are a few things that make voice prompting work smoothly, though:

  • The dictation has to work in every window: browser chat, editor sidebar, terminal, ticket system.
  • It should be triggered by a key, not by clicking a microphone button somewhere.
  • Technical terms, function names and product names must come out right.
  • You should be able to clean up a messy dictation before you send it.

Set up voice to text for AI prompting on Windows

The following steps use SpeVoc, a Windows tool that turns your voice into text in any program you can type in. Offline dictation is free and needs no account, so you can try the whole workflow without signing up anywhere.

Step 1: Install SpeVoc

  1. Open the SpeVoc download page and follow the link to the Microsoft Store.
  2. Click Install. The app is certified and signed by Microsoft, so there is no security prompt.
  3. Start SpeVoc. It runs in the system tray at the bottom right of your taskbar. If you don’t see the icon, click the small arrow (∧) next to the clock.

Requirements: Windows 10 (version 2004 or later) or Windows 11, 64-bit, and a microphone. For offline dictation SpeVoc downloads a speech model on first start (1.5 to 3 GB). A dedicated graphics card speeds things up, but it is not required.

Step 2: Pick a hotkey you can hold without thinking

Out of the box, you hold ↓ + → (both arrow keys) to record. That works right away, but for daily prompting a single key is more comfortable. Many people switch to F9:

  1. Right-click the SpeVoc tray icon and choose Settings….
  2. Open the Recognition tab and click into the hotkey field.
  3. Press the key you want, for example F9. The change applies immediately.

A middle mouse button or side buttons (button 4 or 5) work too, which is handy if one hand is on the mouse anyway. Avoid Ctrl + Alt: on many keyboard layouts it acts as AltGr and blocks characters like @, { or [, which you definitely need when prompting about code.

Step 3: Set the spoken language

In the same Recognition tab, set the language to English if you prompt in English. A fixed language is noticeably more accurate than automatic detection. Available languages include English, German, French, Spanish, Italian, Dutch, Polish and Russian.

Step 4: Teach it your technical vocabulary

Speech recognition struggles with words it rarely hears: framework names, internal project names, abbreviations. SpeVoc has two tools for that in the Dictionary tab of the settings:

  • Custom words help the recognition „hear“ proper names and technical terms correctly.
  • Replacements hard-correct what still comes out wrong. One rule per line: the target word, then the wrong spellings separated by commas, for example Auth-Code: OutCode, out code.

Step 5: Dictate your first prompt

  1. Click into the input field of your AI assistant, whether that is a chat in the browser or an assistant panel in your editor.
  2. Hold your hotkey and talk. The tray icon turns red while recording and yellow while processing.
  3. Let go of the key. When the icon is green again, your text is in the field.

SpeVoc inserts the text via the clipboard, but it backs up whatever you had copied before and restores it afterwards. So if you copied a stack trace to paste into the prompt, it is still there.

A simple structure for spoken prompts

Speaking freely is great, but a loose structure in your head helps the AI and you. This pattern works well:

  1. Goal: „I want the import script to skip rows with an empty email.“
  2. Context: „It’s Python, the file is a CSV with about twenty columns, the email is column four.“
  3. What happens now: „Right now it crashes with a KeyError on the first empty row.“
  4. What you tried: „I added a check for None, but the value is actually an empty string.“
  5. What you want back: „Give me just the changed function, no explanation.“

That takes about thirty seconds to say. Paste the error message or the code snippet with Ctrl+V afterwards, and you have a prompt that would have taken several minutes to type.

Clean up before you send: dictate first, polish second

Raw dictation contains „uh“, repeated words and half sentences. For most AI assistants that is fine. When the text matters more, for example a prompt you want to save as a template, or a message that goes to a person and not to a machine, you can polish it with AI (this needs the online mode and a plan, more on that below):

  1. Dictate the whole text without automatic correction.
  2. Select it and press Ctrl + your hotkey. A window opens with style selection and a preview, for example Standard, Fluent or Business.
  3. Click Insert to replace the selection, or Copy if you can’t edit the source.

If you don’t need the preview, Ctrl + Shift + your hotkey replaces the selection directly. Not happy with the result? Ctrl+Z undoes it. The SpeVoc documentation recommends this „dictate first, style afterwards“ order for longer texts, because the AI then sees the whole text in context instead of sentence by sentence.

Need to fix a single word? Select it, hold the hotkey and say the new word. SpeVoc replaces just the selection, without adding a capital letter, a period or a space.

Offline or online: which mode fits?

SpeVoc has two recognition modes, and both are useful for prompting:

  • Offline: recognition runs with a local Whisper model on your PC. It is free, unlimited, and no recording leaves your computer. Without a GPU, expect roughly 5 to 15 seconds per dictation; with a dedicated graphics card it is considerably faster.
  • Online: cloud transcription in about half a second, plus AI text improvement, styles and translation. It needs an internet connection and a plan with a monthly quota. A free account comes with 10 minutes of trial quota.

One detail for heavy prompt users in online mode: each request is billed with at least 10 seconds. If you fire off lots of tiny dictations, combine your thoughts into one longer prompt. That is a better prompt anyway.

Tips for everyday voice prompting

  • Use a headset if you work in a noisy room. Desk microphones pick up background noise, and in long pauses the recognition can occasionally „hear“ a word that nobody said.
  • Turn off the beep in an open-plan office: the Audio tab in the settings has an option to switch off the signal tone. The colored tray icon still shows what’s happening.

FAQ

Does voice prompting work in any AI tool?

SpeVoc works in any Windows program you can type in: browser, editor, terminal, chat window or ticket system. The text appears wherever your cursor is.

Do I need an account to try it?

No. Download and offline dictation work without registration or credit card. You only create a free account if you want to try online dictation and AI text improvement.

Does my voice leave my PC?

In offline mode, no recording leaves your PC. In online mode, the recording or the selected text is sent to a cloud service for processing.

Can I dictate in English and get German text?

Yes, with the online mode: select the text, press Ctrl + your hotkey and enable translation into up to three target languages at once.

Try it with your next prompt

The fastest way to see whether voice to text for AI prompting suits you is to try it on a real task. Install SpeVoc, set F9 as your hotkey and speak your next prompt instead of typing it. Offline dictation is free forever.

Get SpeVoc from the Microsoft Store or compare the plans for online mode.