Guide

Speech Recognition for Mac

Everything you need to know about using voice recognition and speech to text on your Mac. We'll cover how it works, what's built in, and why local AI is changing the game.

How Speech Recognition Works on Mac

Speech recognition converts spoken words into text. On Mac, there are two fundamentally different approaches: cloud-based and on-device.

Cloud-based recognition sends your audio to remote servers — Apple's, Google's, or someone else's — where powerful hardware transcribes it and sends the text back. This can be very accurate, but your voice data leaves your Mac.

On-device recognition runs the entire AI model locally on your Mac's hardware. Nothing leaves your machine. With Apple Silicon (M1 and newer), Macs now have enough processing power to run state-of-the-art speech recognition models locally with excellent speed and accuracy.

Apple's Built-in Speech Recognition

Every Mac ships with dictation built in. Here's what you get:

• Basic Dictation — On-device, works offline, available in System Settings > Keyboard > Dictation. Limited accuracy and no vocabulary learning.
• Enhanced Dictation — Better accuracy, but sends your audio to Apple's servers for processing. Requires an internet connection and an Apple ID.

To use Apple's built-in speech recognition, press Fn twice (or your configured dictation shortcut) in any text field. Your Mac will listen for a few seconds and insert the transcribed text.

The Local AI Alternative

Modern AI speech recognition models like Whisper (originally developed by OpenAI) can run entirely on your Mac. The model is downloaded once and runs locally — no internet needed, no audio sent anywhere.

This is the approach Arugula takes. It runs whisper.cpp, a high-performance C++ implementation of Whisper, directly on your Mac's Apple Silicon hardware. You get cloud-quality accuracy with complete on-device privacy.

The key advantages of local AI speech recognition:

• Privacy — Your voice never leaves your Mac. No servers, no data collection, no telemetry.
• Speed — No network round-trip. Transcription happens in under a second on Apple Silicon.
• Reliability — Works offline, on airplanes, in areas with no connectivity.
• Vocabulary learning — Apps like Arugula can learn your corrections and improve over time, something cloud services can't easily do per-user.

Comparing Your Options

For a detailed feature-by-feature breakdown, see our full comparison.

What About MacBook Air and MacBook Pro?

All Apple Silicon MacBooks — including MacBook Air (M1, M2, M3) and MacBook Pro (M1 Pro, M2 Pro, M3 Pro, M4 Pro, and Max) — have more than enough processing power for local speech recognition. The AI model runs efficiently on the Neural Engine built into every Apple Silicon chip.

MacBook Air users will see slightly longer initial model load times compared to Pro models, but ongoing transcription speed is essentially the same. Battery impact is minimal — Arugula only uses processing power when you're actively dictating.

Speech Recognition for Specific Workflows

Voice recognition on Mac isn't just for writing documents. People use it for:

• Email — Dictate replies and compose messages in Gmail, Apple Mail, or Outlook
• Messaging — Quick voice replies in Slack, Discord, and Messages
• Code comments — Document your code without leaving the editor
• Meeting notes — Capture key points during video calls
• Accessibility — Essential for users with RSI, arthritis, or mobility limitations

See our full guide on voice typing use cases for more details.

Getting Started with Speech Recognition on Mac

The fastest way to try voice recognition on your Mac:

1. Try Apple's built-in — Go to System Settings > Keyboard > Dictation and turn it on. Press Fn twice to start.
2. Try Arugula for better accuracy and privacy — Download Arugula, install it, and hold Right Option to dictate. That's it.

For a more detailed walkthrough, see our how to voice type on Mac guide.

System Requirements

For the best speech recognition experience on Mac:

• macOS 15 (Sequoia) or later — required for Arugula
• Apple Silicon (M1 or newer) — required for local AI speech recognition at usable speeds
• Microphone — built-in is fine for quiet environments; external mic recommended for noisy spaces