macOS Tahoe replaced the older speech recognition engine with Apple's foundation model, running entirely on your Apple Silicon chip. If you're a developer who spends most of the day in a text editor or terminal, this matters.

What changed in macOS Tahoe

The new dictation engine is a significant improvement:

  • Better general accuracy — improved for everyday speech and common tech terms like "React," "API," or "TypeScript"
  • Fully on-device — audio stays on your Mac, no network dependency
  • Lower latency — no round-trip to a server
  • Smarter punctuation — the model infers punctuation from speech cadence
  • Free — ships with the OS, improves with hardware generations

Setting up built-in Dictation

Step 1: Enable Dictation

  1. Open System Settings
  2. Click Keyboard in the sidebar
  3. Scroll to Dictation and toggle it on
  4. Enable Auto-punctuation

Step 2: Pick your hotkey

In System Settings > Keyboard > Dictation, choose a shortcut. Control Key Twice works well — it's fast and doesn't interfere with IDE shortcuts.

Step 3: Start talking

  1. Click into any text field — VS Code, Terminal, browser, Slack
  2. Press your hotkey
  3. Speak naturally as text appears
  4. Press the hotkey again or click Done

Using voice in developer workflows

Dictating prompts to AI coding tools

Voice input enables longer, more detailed prompts. You can dictate a 200-word prompt in about 30 seconds — making it practical to include full context, constraints, and requirements instead of compressing everything into a single sentence.

Writing Slack messages and documentation

Explanatory messages where clarity matters more than brevity. Instead of spending two minutes typing a paragraph in Slack, speak it in 20 seconds.

Code review comments

Dictation helps articulate reasoning behind suggested changes. "I think we should move this validation to the service layer because right now it's duplicated in three controllers" is easier to say than to type.

Capturing ideas without losing context

When you're deep in code and an idea for a different part of the system comes up, dictate a quick note instead of switching context.

Tips from daily use

  • Speak normally. The foundation model was trained on natural speech. Over-enunciation reduces accuracy.
  • Add tricky words to Text Replacements. Custom product names and library terms can be pre-mapped via System Settings > Keyboard > Text Replacements.
  • Don't watch the words appear. Real-time display causes mid-sentence second-guessing that disrupts flow.
  • Use a headset mic in noisy spaces. Built-in laptop mics pick up ambient noise that degrades accuracy.

Where built-in Dictation falls short

The built-in engine handles general speech well, but developers hit its limits quickly:

  • Programming vocabulary — library names, CLI commands, variable names, and domain-specific jargon get mangled regularly
  • No post-processing — output cannot be reformatted, cleaned up, or transformed before pasting
  • No transcript history — no searchable log of what you dictated
  • Short bursts only — not designed for meetings or extended recording sessions
  • No translation — single-language output only

Going beyond built-in Dictation with Vext

Vext addresses each of these limitations:

Faster transcription

Vext uses the Parakeet engine via CoreML, running at 150x realtime on Apple Silicon — a 60-second recording processes in under half a second. Apple's built-in Dictation runs at approximately 25x realtime.

Enhance

AI post-processing that cleans up filler words, fixes sentence structure, and smooths spoken language into polished text. Runs locally on your Mac through models like Gemma 3 4B.

Live translation

Speak in any language, get text in your target language. When combined with Enhance, cleanup and translation happen in a single pass.

Meeting transcription

Record full meetings with speaker identification, AI summaries, and screenshot capture. Works with Zoom, Google Meet, FaceTime, and any audio source.

Voice notes

Quick voice memos stored locally in the app. Same processing pipeline as dictation — just saved for later instead of pasted at your cursor.

YOLO Mode

Auto-submit prompts to AI coding tools. Speak, release, and your prompt is already running in Claude Code or ChatGPT.

Three transcription engines

Choose between Parakeet (fastest, local), Apple Dictation (built-in), or OpenAI-compatible APIs. Switch based on your needs.

Getting started with Vext

brew install muvon/tap/vext

Free trial: 100 dictations, 50 notes, 10 meeting recordings. No account required.

The built-in macOS Dictation is a solid starting point. When you hit its limits — and in development workflows, you will — Vext picks up where Apple leaves off.