Speech to text on a Mac used to mean one of two things: Apple's built-in dictation or Dragon NaturallySpeaking. Apple's version was limited but free. Dragon was powerful but expensive, and then Nuance killed the Mac version entirely. For years after that, the options were bleak.
That's no longer the case. Apple Silicon changed the game by putting a Neural Engine in every Mac — dedicated hardware for running machine learning models locally. A new wave of speech-to-text tools has emerged that takes advantage of this hardware, and some of them are genuinely good. Here's a practical breakdown of what works, what doesn't, and what to use depending on your needs.
Apple Dictation: free but frustrating
Apple's built-in dictation is the obvious starting point. It's free, it's pre-installed, and on Apple Silicon Macs it runs partially on-device. Go to System Settings > Keyboard > Dictation, toggle it on, and double-tap the Fn key to start speaking.
For short dictation — a quick search, a one-line reply — it works well enough. But the problems emerge fast once you try to use it for real work:
- Timeout. Apple Dictation stops listening after a brief pause. If you stop to think mid-sentence, it cuts off and you have to restart.
- No cleanup. Every "um," "uh," and "you know" goes straight into your text. Real speech is messy, and Apple does nothing to clean it up.
- Inconsistent accuracy. Short phrases transcribe fine. Longer passages accumulate errors — wrong homophones, missed words, garbled punctuation.
- App compatibility. It doesn't work reliably in terminal apps, some code editors, or Electron-based apps. If you live in VS Code or use a terminal regularly, you'll hit walls.
Apple Dictation is a baseline, not a solution. It proves speech-to-text works on your Mac. It just doesn't work well enough for the people who need it most. For a deeper look at what you're missing, see our breakdown of how to dictate on Mac without Siri.
Cloud tools: powerful but problematic
Cloud-based speech-to-text apps send your audio to remote servers for processing. The appeal is straightforward: bigger models, more compute power, more features. Tools like Wispr Flow and Otter.ai fall into this category.
Wispr Flow is the most polished of the bunch. It transcribes speech and can rewrite output to match the context of the app you're typing in — more formal in email, more casual in Slack. Otter.ai focuses on meeting transcription and offers AI-generated summaries.
The tradeoffs are real, though:
- Privacy. Your voice — every word you dictate — travels to a server you don't control. If you're dictating client emails, medical notes, legal strategy, or anything confidential, this is a serious issue.
- Latency. Even on fast connections, the upload-process-download cycle adds noticeable delay. On slower connections, it's worse.
- Cost. Wispr Flow is $10/month. Otter.ai's useful features are behind a paywall too. At $120/year, the recurring cost adds up.
- Dependency. No internet means no speech-to-text. Planes, spotty Wi-Fi, rural areas, government buildings with restricted networks — cloud tools don't work when you need them most.
For teams that need meeting transcription and don't handle sensitive data, cloud tools can make sense. For individual dictation — the "hold a key and talk" workflow — the privacy and cost tradeoffs rarely justify the benefits.
Local speech-to-text: the new default
The most significant shift in speech-to-text on Mac is the move to fully local processing. Thanks to Apple Silicon's Neural Engine and frameworks like CoreML, speech recognition models can now run entirely on your Mac with quality that rivals cloud services.
Several apps have emerged in this category. Here's how they differ:
SuperWhisper
Built on OpenAI's Whisper model, SuperWhisper offers local transcription with multiple model sizes. Larger models are more accurate but slower. It's a capable tool with a solid community. The subscription pricing ($10/month or $100/year) and the need to configure model sizes make it better suited for technical users who want fine-grained control.
MacWhisper
Also Whisper-based, but designed primarily for file transcription rather than live dictation. Excellent for transcribing recorded meetings, interviews, and voice memos. Not ideal if what you want is system-wide "talk and it types" dictation.
Voiced
Voiced is a speech-to-text app for Mac built for live dictation. It uses Apple's CoreML framework to process speech on-device, so there's no cloud dependency, no model configuration, and no account required. The key differentiators:
- Works in any app. Mail, Slack, Notion, Google Docs, VS Code, Terminal — anywhere you can type, you can dictate. This matters because the whole point of speech-to-text is speed, and switching between apps to copy-paste text defeats the purpose.
- Smart Cleanup. Voiced automatically removes filler words and polishes your speech. You say "So um I think we should uh probably move the deadline to Friday." You get "I think we should move the deadline to Friday." No manual editing required.
- No timeout. Hold your hotkey and talk for as long as you need. Pause to think. It keeps listening until you release. This alone makes it usable for the kind of longer dictation that Apple's version can't handle.
- $40 one-time. No subscription. No recurring charges. That's less than four months of a cloud alternative.
For a full comparison of all the options, see our guide to the best voice-to-text apps for Mac in 2026.
Which approach should you use?
It depends on what you're doing:
- Quick one-liners: Apple Dictation is fine. It's free and already on your Mac.
- Meeting transcription: Otter.ai or MacWhisper, depending on whether you want cloud AI features or local privacy.
- Daily dictation for emails, writing, and AI prompts: A local app like Voiced. You get the speed, the privacy, the reliability, and you pay once. If you're using AI coding tools, voice dictation pairs especially well with Cursor, Claude Code, and similar tools.
The speed advantage is real: most people speak at 150 words per minute but type at 80. Switching to voice for your first drafts, emails, and messages can genuinely cut your writing time in half. The only question is which tool you use to get there.