How to Dictate Code & Comments in VS Code (Zero Latency)
Developer Tools 5 min read •

How to Dictate Code & Comments in VS Code (Zero Latency)

Boost your developer productivity by dictating comments, git commits, and documentation directly into VS Code using local AI speech-to-text.

The Context Switching Problem

As developers, we spend about 50% of our time reading and writing code, but a significant chunk is spent on "human text": code comments, documentation, git commit messages, and pull request descriptions.

Typing these out breaks your flow. You switch from "logic mode" (thinking in variables and functions) to "prose mode". This context switch is costly. Voice dictation bridges this gap, allowing you to dump your thoughts into the editor at 150 words per minute.

Why Not Use Windows Dictation?

Windows has built-in dictation ( Win+H), but it has three major problems for developers:

  • Latency:It often lags as it sends audio to the cloud.
  • Focus Stealing:The UI pops up and can steal focus from your cursor position.
  • Privacy:Do you really want to send proprietary code snippets to Microsoft servers?

The Local Workflow

Parakeet Flow solves this by running a compact, high-performance model (Parakeet V3 via Sherpa-ONNX) directly on your machine. Here is the ideal workflow for VS Code:

Step-by-Step Setup:

  1. 1. Install Parakeet Flow and launch it. It runs quietly in the system tray.
  2. 2. Open VS Code and place your cursor where you want to write a comment.
  3. 3. Hold the global hotkey (Default: Ctrl+Space).
  4. 4. Speak your comment. Release the keys.
  5. 5. The text is instantly pasted at your cursor position.

Best Use Cases

Git Commits

"Refactor the authentication middleware to use the new token provider and fix the race condition in the login flow."

Complex Logic Explanation

"TODO: This function uses a naive exponential backoff. We should replace it with a jittered approach if we hit scale limits."

Quick Notes & TODOs

Capture fleeting thoughts like "Note to self: refactor this when we add caching" before they slip away.

Markdown Docs

Writing `README.md` files becomes much faster when you can just talk through the features.

Optimizing for Code

While the model is trained on general speech, it handles technical jargon surprisingly well because the training data includes vast amounts of internet text (including technical discussions).

Pro Tip:You can speak punctuation. Saying "Open bracket user id equals null close bracket" will often be transcribed correctly, though for heavy syntax, it's usually faster to type the symbols and dictate the variable names/comments.

← Back to Blog