Speech-to-Text Apps That Work Completely Offline

Yes, speech to text works without internet, as long as the app keeps its speech recognition model on your device. You download the model or language pack once. After that, recording and transcription run on your phone’s or computer’s own processor, in airplane mode, on a plane or in a basement. The best offline options by device:

• iPhone: built-in dictation (in on-device languages), Private Transcribe, Whisper Notes and Aiko.
• Android: Pixel Recorder, Live Transcribe’s offline mode, FUTO Voice Input and Private Transcribe.
• Mac: Apple Dictation, MacWhisper and VoiceInk.
• Windows: Voice Access, Buzz, Vibe and Handy. Windows voice typing (Win + H) is not offline.
• Linux: Speech Note, Buzz and whisper.cpp.

The catch is that plenty of apps sold as “offline” still need a connection for some features, so test one before you rely on it.

Offline speech-to-text apps at a glance #

Prices are from each developer’s site in September 2026.

How does speech recognition work without internet? #

Speech-to-text needs a trained model, a large file of learned patterns that maps sounds to words. Cloud apps keep that model on their servers and send your audio to it. Offline apps put the model on your device instead.

Open models made this practical. OpenAI’s Whisper, and the whisper.cpp project that runs it on ordinary hardware, fit a capable model into a few hundred megabytes. Phone-sized Whisper models run from under 100 MB for quick notes to a little over 500 MB for the most accurate ones. Desktop apps can use larger versions if you have the memory: OpenAI lists its largest Whisper model as needing about 10 GB of video memory.

Once the model is on your device, the internet has nothing left to contribute to the transcription itself.

Offline speech-to-text on phones #

Private Transcribe (iPhone and Android) #

Private Transcribe is our app, and it’s built around this idea. It runs Whisper through whisper.cpp on the phone’s own processor. There’s no account and no transcription server, and after a one-time model download it works with networking switched off. You pick from five model tiers, from Quick (78 MB) to Best (574 MB, Whisper large-v3-turbo), and the multilingual models cover 99+ languages.

You record, or import an audio or video file (m4a, mp3, wav, aac, mp4 or mov, up to 90 minutes). Transcription starts when you stop, and the text lands in a searchable history with a timestamps view. Custom vocabulary keeps names and jargon spelled your way.

What it doesn’t do: it won’t show words while you’re still talking, it doesn’t label speakers, and it doesn’t type into other apps the way a keyboard does. It imports m4a, mp3, wav and aac audio and mp4 or mov video, and its decoder also reads .ogg and .opus files, the format WhatsApp uses for voice notes. Older .amr or .3gp recordings may need converting to m4a or mp3 first. The free version shows AdMob ads, which use the network but never see your recordings. Pro is a one-time $7.99 in the US that removes ads and adds .txt and .srt export plus translation into English. Use a multilingual Quick, Balanced, Accurate or Professional model for translation. The Best tier’s large-v3-turbo model wasn’t trained for translation, so don’t count on it there.

Apple Dictation (iPhone) #

Tap the keyboard microphone in any app. In supported languages it runs on the device, and Apple says Settings > General > Keyboard shows whether yours does. It’s the fastest way to write a message by voice, but it only handles live speech and doesn’t keep the audio.

Whisper Notes and Aiko (iPhone, iPad and Mac) #

Whisper Notes runs on iPhone 12 and later, iPad and Apple silicon Macs. It offers Parakeet, Whisper Large V3 Turbo and SenseVoice engines, labels speakers with timestamps for each turn, and costs $7.99 one-time on iOS. Aiko, from Sindre Sorhus, transcribes recordings with Whisper’s medium or small model on iPhone and exports text and subtitles. More iPhone options are compared in offline dictation apps for iPhone.

Pixel Recorder (Pixel phones) #

Recorder transcribes on the device as you record on Pixel 3 and later, with speaker labels on Pixel 6 and later. Two caveats from Google’s help page: the re-transcribe option may process audio on Google’s servers, and AI summaries run in the cloud. “Back up & sync” is optional and off unless you turn it on. Recorder also can’t transcribe audio files from other apps.

Live Transcribe and FUTO Voice Input (Android) #

Live Transcribe gives live captions offline, but only for languages you’ve downloaded, and some phones on Android 8 to 11 can download only English, per Google. It keeps caption history for three days. FUTO Voice Input types by voice on the device. The Android options are compared in private voice to text on Android without internet.

Offline speech-to-text on computers #

Mac. Apple Dictation handles everyday dictation on the device in supported languages. For recorded files, MacWhisper’s free version already includes speaker recognition and subtitle export. For hotkey dictation with bigger models, VoiceInk and Whisper Notes run locally.

Windows. Voice Access in Windows 11 works without an internet connection. Windows voice typing (Win + H) is a different feature. Microsoft says it uses online speech recognition powered by Azure, so your speech leaves the PC. Buzz and Vibe transcribe files with Whisper, and Handy does push-to-talk dictation. The full Windows list is in best offline speech recognition apps for Windows.

Linux. Speech Note (from Flathub) does offline speech-to-text, text-to-speech and translation. Buzz and whisper.cpp handle files and batch jobs, and whisper.cpp’s whisper-stream gives live captions.

What still needs a connection in “offline” apps? #

This is where people get caught out:

• The first download. Every offline app has to fetch its model or language pack once.
• Only some languages. Live Transcribe works offline only in languages you’ve downloaded, and Apple Dictation only in its on-device languages.
• Only some features. Pixel Recorder’s re-transcribe option and its summaries can use Google’s servers.
• Look-alike tools. Windows voice typing needs Azure. Voice Access doesn’t.
• Mixed apps. Superwhisper and Spokenly have good local models but also cloud features. Use Spokenly’s Local Only Mode, and pick local models in Superwhisper.
• Cloud-only tools. Wispr Flow states that “transcription always happens in the cloud.” Speechnotes hands dictation to the browser’s or operating system’s speech engine. Otter and most meeting assistants transcribe on their servers.
• The extras. Ads, purchase checks, crash reports and sync use the network even when transcription doesn’t. In Private Transcribe, the network is used only for the model download, ads in the free version, and Apple’s or Google’s check when you buy Pro.

How to test whether an app is really offline #

1. Open the app while connected and download any model or language it asks for.
2. Turn on airplane mode and check that Wi-Fi is off too. Airplane mode can leave Wi-Fi on if you enabled it before. On a computer, disconnect from the network.
3. Dictate, record ten seconds of speech, or import a short clip.
4. Watch what happens. Text means recognition happens on the device. An error, a spinner or “check your connection” means it relies on a server.

It takes under a minute and tells you more than any product page. If the app passes, read its App Store or Play Store privacy label to see what else it collects when you’re online.

Is offline transcription as accurate as online? #

For clear speech from one person, the difference is small enough that most people won’t notice it. Cloud services still have an edge with heavy background noise, many overlapping speakers and less common languages, where very large server-side models and extra processing steps help. Offline, accuracy depends mostly on the model size you pick. Larger models are slower and use more battery, but they don’t need a connection either way.

What “offline” doesn’t protect you from #

On-device transcription stops one kind of exposure: a vendor holding your audio. It doesn’t stop the others.

• Sync and backups. If your phone backs up app data to the cloud, or you save transcripts to a synced folder, copies leave the device.
• Copy and paste. Pasting a transcript into a cloud note app puts the text there.
• The phone itself. Anyone who can unlock it can read your notes, so use a strong passcode.

For step-by-step setup on each platform, see how to transcribe voice notes to text offline.

Which offline app should you choose? #

• Typing by voice on your phone: your keyboard’s built-in dictation, confirmed on-device.
• Voice notes, lectures and meetings on a phone: Private Transcribe on iPhone or Android, or Recorder on a Pixel.
• Live captions on Android: Live Transcribe in offline mode.
• Conversations with speaker names on Apple devices: Whisper Notes.
• Files with speaker labels on a computer: MacWhisper on Mac, Buzz on Windows or Linux.
• Free and open source: Buzz, Vibe, Handy, whisper.cpp or Speech Note.

Frequently asked questions #

Does speech to text work without an internet connection? #

Yes, if the speech model is stored on your device. Built-in dictation in on-device languages, Pixel Recorder, Live Transcribe’s offline mode and Whisper-based apps all work in airplane mode once their models are downloaded. Cloud tools such as Windows voice typing and Wispr Flow stop working without a connection.

What is the best free offline speech-to-text app? #

On a phone, your built-in dictation is free, and Private Transcribe is free with ads. On a computer, Buzz, Vibe, Handy and whisper.cpp are free and open source, and MacWhisper has a capable free version for Mac.

Which offline speech-to-text app is the most accurate? #

Accuracy depends mostly on the model size. Apps that run Whisper large-v3-turbo or larger, such as MacWhisper, Whisper Notes, Buzz and Private Transcribe’s Best tier, are generally the most accurate offline, especially with accents and background noise.

How much storage do offline speech models need? #

From under 100 MB for small, fast models to over 1 GB for the largest desktop ones. Private Transcribe’s five tiers run from 78 MB to 574 MB, and most people keep one or two. Mid-size models of about 150 to 200 MB suit most phones.

Which private voice note app is best for Android? #

Pixel Recorder, if you have a Pixel and only need its own recordings. Private Transcribe runs on any recent Android phone, imports files and never uploads audio, though it has no speaker labels and shows ads unless you buy Pro.