Voice to Text Offline: Type by Voice in Any App

Hold Fn in any Mac app, speak, and let go. The words appear at your cursor, transcribed on your Mac with no internet connection.

How does offline voice typing work?

Whisper Notes types what you say into whichever app has your cursor. Hold the Fn key, speak, and release: the speech model on your Mac turns the audio into text and types it at the cursor. Nothing is uploaded, so it works with Wi-Fi off. Parakeet V3 is the default engine; Whisper Large V3 Turbo covers 100+ languages, and SenseVoice covers Chinese, Japanese, Korean and Cantonese.

•Mac: hold Fn to dictate into any app — Claude, ChatGPT, Slack, VS Code, Mail
•iPhone: one tap on the Lock Screen widget opens the app and starts recording
•Everything runs on your device; nothing is uploaded
•iPhone: $7.99 once. Mac: 5,000 words a week free, then $14 once. No subscription.

Mac (macOS 14+, Apple Silicon) · iPhone (iOS 18+)

Looking to transcribe recorded audio files instead of dictating? See our offline transcription guide.

On Mac: Dictate Into Any App

Anywhere you can type on your Mac, you can speak instead: email drafts, Slack replies, code comments, AI prompts. Hold Fn, talk, release, and the words appear at the cursor. This is dictation, not file transcription, and it works with Wi-Fi off. Voice typing is in the Mac app downloaded from our website; Apple's rules keep it out of the Mac App Store version.

System-Wide Voice Input

Whisper Notes installs a global shortcut. In any app—Claude, ChatGPT, Gemini, Gmail, Notion, VS Code, even Terminal—hold Fn and speak. When you release, the on-device model — Parakeet V3 by default, with Whisper Large-v3 Turbo and SenseVoice also available — processes your audio locally. Text appears where your cursor is.

•Works in any app where you can type
•Text appears at the cursor
•Parakeet V3 by default (~60× realtime on Apple Silicon); Whisper Large-v3 Turbo and SenseVoice also run locally
•Turn it on in the left side panel: Voice Typing → Enable voice typing

Works in: Claude, ChatGPT, Gemini, Gmail, Slack, VS Code, Terminal, Notion

Files, Too

The same app transcribes audio and video files. Import a file and the transcript fills in as each part finishes, so you can start reading before the end.

Custom Vocabulary

Speech models stumble on names and jargon. Add your vocabulary—company names, product names, technical terms—and capitalization is preserved ("Claude Opus 4.5" stays "Claude Opus 4.5", not "claude opus").

Claude, GPT-4, Gemini, Whisper · OAuth, TypeScript, Kubernetes · HIPAA, GDPR, SOC2

Silence Handling

Whisper-family models hallucinate on long silences—repeating phrases or inventing words, because the decoder is built to always emit something. Voice Activity Detection finds the speech and trims the dead air before transcription, so there is far less silence left to fill. It reduces the problem rather than removing it; recording with clean, natural pauses helps too.

How Fast?

M4 Pro: A 20-second dictation appears in about 2.5 seconds. On files, Parakeet V3 runs at 103× real-time and Whisper Large-v3 Turbo at 14.3×

M5 Air: ~60× real-time with Parakeet V3 — a 17.5-minute meeting came back in 16.7 seconds

M1 and later: Every Apple Silicon Mac runs all three engines. Older chips are slower, still well past real-time

Dictating to AI Assistants

Prompting Claude

Hold Fn, describe your problem in detail, release, send. No copy-paste from a separate app.

Slack and Email

Hold Fn in the compose field, say what you mean, release.

First Drafts

Some writers find dictated first drafts looser than typed ones. Get the ideas out first, edit later.

On iPhone: Record From the Lock Screen

iPhone has no Fn key, so Whisper Notes works differently there: it records and transcribes on the phone, and you copy or share the text. One tap on the Lock Screen widget opens the app and starts recording.

Lock Screen Widget

•One tap opens the app and starts recording
•A Live Activity shows the duration while you speak
•The Dynamic Island shows the recording status

Voice Memos and Files

•Share a recording from Voice Memos into Whisper Notes
•Import audio files from the Files app
•Transcription runs on the iPhone itself

Export Anywhere

•Copy to the clipboard
•Share to Notes, Messages, email, any app
•Export with timestamps for review
•SRT format for video subtitles

What People Record

Before Sleep

"That API design is wrong. Events should be immutable. Refactor to event sourcing pattern first thing tomorrow."

Morning Run

"Article idea: the keyboard as a thought compression algorithm. We write differently than we think because typing is slow."

Walking

"The meeting is stuck because we're optimizing the wrong metric. Reframe around retention, not engagement."

Why Offline Matters

Your Audio Never Leaves Your Device

•No server upload: processing happens on the iPhone or the Mac itself
•No retention policy to read, because nothing is sent anywhere
•Confidential conversations and work notes stay on hardware you own
•We don't collect or upload any user information

Works Without Internet

•Airplane mode, subway tunnels, spotty Wi-Fi—all work
•Secure facilities that block network access—works
•Latency is just processing time, not network round-trip
•Performance doesn't degrade when servers are overloaded

Pay Once

•iPhone: $7.99 once. Mac: 5,000 words a week free, then $14 once.
•No per-minute charges and no subscription
•Cloud dictation is billed monthly: Wispr Flow Pro is $15 a month or $144 a year

How It Compares

Feature	Whisper Notes	Apple Dictation	SuperWhisper	Wispr Flow
Processing	100% on-device	On-device for many languages	Local or cloud, by mode	Cloud servers
Platforms	Mac, iPhone	Mac, iPhone	Mac, Windows, iPhone	Mac, Windows, iPhone
Lock Screen Widget	Yes	No	No	No
Dictation in any app	Yes (Mac, hold Fn)	Yes	Yes	Yes
Price	$7.99 iPhone / $14 Mac, once	Free	$8.49/mo, $84.99/yr or $249.99 lifetime	$15/mo or $144/yr
Speech Models	Parakeet V3 + Whisper + SenseVoice	Apple's own	Whisper variants + Parakeet	Cloud models
Custom Vocabulary	Yes	No	Yes	Yes

Apple Dictation is free and good enough for short phrases. Superwhisper and Wispr Flow add AI rewriting for a subscription, and Wispr Flow runs in the cloud. Whisper Notes runs on the device for a one-time price. Competitor prices were checked on their sites; check them for current pricing.

Trade-offs

Local processing has real trade-offs:

Model Download

The default engine is Parakeet V3, a 465 MB model that runs on all Apple Silicon Macs at roughly 60× realtime, for English and 24 other European languages. For 100+ languages, download Whisper: Large V3 Turbo (~1.6 GB) on Mac, Whisper Small on iPhone. For Chinese, Japanese, Korean and Cantonese, select SenseVoice.

Apple Only

This is an Apple app: an Apple Silicon Mac (M1 or newer) with macOS 14, or iPhone 12 or later with iOS 18. There is no Android, Windows or Intel Mac version.

Speed vs Cloud

Local inference is bounded by your device rather than a datacenter. On iPhone, the App Store listing reports about 18 seconds for five minutes of audio with Parakeet on an iPhone 15, and about a minute with Whisper — so an hour-long recording is minutes, not seconds. If you need an hour-long file back instantly, a cloud service is still faster.

Accuracy Ceiling

On clear speech, the error rate is in the single digits. Heavy accents, loud background noise, or mumbling will need some editing. If you need 99.9% accuracy for a medical transcript, hire a human transcriptionist. If you need a usable draft quickly and privately, this works.

Get Started

iPhone

1.Download Whisper Notes from the App Store ($7.99)
2.Open it; Parakeet V3 is built in
3.Long-press the Lock Screen → Customize → Lock Screen → add the Whisper Notes widget
4.Tap the widget to open the app and start recording

Mac

1.Download Whisper Notes for Mac (free for 5,000 words a week)
2.Allow the microphone when asked
3.Allow Accessibility, which the app uses to type at your cursor
4.If you skipped it: open Voice Typing in the left side panel and turn on Enable voice typing
5.Hold Fn anywhere and start talking