You are on a flight, in a coffee shop with unreliable WiFi, or working from a cabin with no cell signal. You need to get text down quickly, and your keyboard feels slow. Should you be locked out of voice dictation just because you are offline? Steno does not think so. That is why it includes a fully functional offline mode that works without any internet connection.
How Offline Mode Works
Steno's primary transcription engine is the Groq Whisper API, which delivers sub-second, highly accurate transcriptions via the cloud. But when your Mac has no internet connection, Steno automatically falls back to Apple's on-device speech recognition framework. This is the same technology that powers Siri's on-device processing and Apple's built-in dictation when offline mode is enabled in System Settings.
The switch is seamless. You do not need to change any settings or toggle any modes. Steno detects the network state and routes your audio accordingly. Hold the hotkey, speak, release. The interaction is identical whether you are online or offline. The only difference is what happens behind the scenes.
On-Device Speech Recognition
Apple's Speech framework provides access to speech recognition models that run entirely on your Mac's processor. On Apple Silicon Macs (M1 and later), these models leverage the Neural Engine, a dedicated hardware component designed for machine learning inference. The models are downloaded once and stored locally, so no internet is needed at recognition time.
When Steno uses on-device recognition, your audio never leaves your Mac. It is processed locally, transcribed locally, and the result is inserted at your cursor. The entire pipeline runs within your machine's memory.
Accuracy: Cloud vs. Offline
Honesty matters when discussing offline transcription. The cloud-based Groq Whisper API, which uses the Whisper large-v3 model, is significantly more accurate than Apple's on-device speech recognition. Here is what you can expect from each:
Cloud Mode (Groq Whisper)
- Word error rate of approximately 3-5% for clear English speech
- Excellent handling of technical vocabulary, proper nouns, and uncommon words
- Strong punctuation and capitalization
- Good performance with accented English
- Reliable even in moderately noisy environments
Offline Mode (Apple On-Device)
- Word error rate of approximately 8-15% depending on vocabulary and speaking style
- Good with common words and phrases, weaker with technical jargon
- Basic punctuation; capitalization is sometimes inconsistent
- Best with standard American English; performance varies with other accents
- More sensitive to background noise
The accuracy gap is real but manageable for most dictation tasks. If you are composing an email, taking quick notes, or jotting down ideas, offline mode handles these well. If you are dictating a legal brief with precise terminology or transcribing a medical report, you will want the cloud mode's superior accuracy.
Speed in Offline Mode
Offline transcription speed depends on your Mac's hardware. On M1 and later chips, Apple's on-device recognition processes speech in near real-time, with results typically appearing within 1-2 seconds after you stop speaking. This is slower than Steno's cloud mode (which completes in under one second) but still fast enough to maintain a productive dictation workflow.
On older Intel Macs, offline processing can take 2-4 seconds for a typical sentence. The experience is usable but noticeably slower. If you regularly work offline and speed is important, an Apple Silicon Mac makes a significant difference.
When to Use Offline Mode
Travel
Flights, trains, and road trips often mean spotty or nonexistent internet. Offline mode ensures you can keep dictating regardless of connectivity. Many Steno users report that flights are some of their most productive writing times because the lack of internet removes distractions, and offline dictation keeps the words flowing.
Unreliable Networks
Coffee shops, conference venues, and hotel WiFi are notorious for inconsistent connections. Rather than dealing with failed transcriptions when the network drops, Steno's automatic fallback to offline mode means you never miss a beat. If the connection comes back, Steno switches back to cloud mode on the next dictation.
Privacy-Sensitive Work
Some work is too sensitive to send over the network, even to a trusted API. If you are dictating confidential information, trade secrets, or personal notes that you do not want leaving your device, offline mode guarantees that your audio is processed entirely on your Mac. No data is transmitted anywhere.
Air-Gapped Environments
Some workplaces, particularly in government, defense, and finance, have machines that are deliberately not connected to the internet. Steno's offline mode works perfectly in these environments, giving users voice dictation capability on machines that have no network access.
Privacy Benefits of Offline Transcription
The privacy advantage of offline mode deserves special attention. When audio is processed on-device, the privacy guarantee is absolute. Your voice data is not transmitted, not stored on any server, and not used to train any model. It exists only in your Mac's memory during processing and is discarded immediately after.
This is a meaningful distinction from cloud-based transcription, even when the cloud provider has strong privacy policies. With offline mode, privacy is not a matter of policy. It is a matter of physics. The data simply never leaves your hardware.
Steno's cloud mode also respects privacy. Audio sent to the Groq API is processed and discarded; it is not stored or used for training. But for users who need the strongest possible privacy guarantee, offline mode provides it.
Setting Up Offline Mode
Steno's offline mode requires no special setup. However, you should ensure that Apple's on-device speech recognition is available on your Mac:
- Open System Settings and navigate to Keyboard settings.
- Under Dictation, ensure that your preferred language is downloaded for on-device use.
- macOS will download the necessary language model, which is typically 100-300MB depending on the language.
Once the language model is downloaded, Steno will automatically use it whenever cloud transcription is unavailable. You can also manually select offline mode in Steno's settings if you prefer to always transcribe locally.
The Best of Both Worlds
Steno's approach gives you the best of both worlds: cloud-grade accuracy and speed when you are online, with reliable offline fallback when you are not. You never have to think about which mode you are in. The app handles the transition automatically, and the interaction model stays exactly the same. Hold the key, speak, release. Text appears.
Steno is available with a free tier and a Pro tier at $4.99/month. Both tiers include offline mode. Download at stenofast.com.
Dictation should work everywhere you do. Offline mode ensures that your voice is never silenced by a missing internet connection.