InkvoxDictation Without Internet: What Actually Works
Your dictation app worked fine on Wi-Fi. Then you boarded a flight, or walked into a signal-dead office, and it just stopped. Here's which tools keep working with zero connection — and which only look offline until you actually need them to be.
Yes, some dictation tools work with absolutely no internet connection — but only one category: apps that run the speech-to-text model directly on your own CPU or GPU, with no server in the loop. Windows' built-in Voice Typing (Win+H) is not one of them, and neither is Wispr Flow — both need an active connection for every single sentence, because the transcription itself happens on a remote server, not your machine. If you're dictating on a plane, in a signal-dead office, on a network with no outbound access, or anywhere a connection isn't guaranteed, the tool that will actually keep working is a local speech-to-text app, with its model already downloaded before you lose the connection — not after.
TL;DR: Cloud dictation (Windows Voice Typing, Wispr Flow) requires internet for every sentence, full stop. Local Whisper-based apps — including Inkvox — run the model on your own hardware and work with the network off, provided the model was downloaded ahead of time. Test that before you need it, not mid-flight.
Why do most dictation tools need an internet connection?
Cloud dictation apps don't run speech recognition on your device at all. They record your voice, stream the audio to a server, run the model there, and stream the text back to your cursor. That round trip is the entire product — which means no connection, no round trip, no text.
Microsoft is explicit about this for its own built-in tool: per its support documentation, "voice typing uses online speech recognition, which is powered by Azure Speech services," and "to use voice typing, you'll need to be connected to the internet." There's no offline setting to flip — it's how the feature is architected. (We cover the full story, including why Windows quietly retired its one genuinely offline dictation tool, in why Windows Voice Typing needs the internet.)
Wispr Flow is built the same way, just as a third-party product instead of an OS feature. Per Wispr's own security documentation, "Wispr Flow runs entirely in the cloud and is delivered as multi-tenant SaaS, hosted with a major US cloud provider." Entirely in the cloud means exactly that: there's no local model to fall back on when the network drops.
"Offline" as a fallback vs. offline by design
Not every dictation tool that struggles offline is dishonest about it — but the way "offline" gets used in this space hides a real architectural difference. Some products are cloud-first with an emergency local mode: a smaller, lower-accuracy model you can download for outages, often missing the AI cleanup, custom vocabulary, or formatting features that work in the normal cloud mode. That's offline as a fallback — a degraded version of the product, and one you may not discover the limits of until you're already offline and it happens.
A tool that's offline by design doesn't have that split. There's no cloud mode to compare against, so there's nothing to degrade into — the model on your device is the only model, and it behaves the same on a plane as it does at your desk. That's the distinction worth checking before you rely on a tool somewhere you can't test it: is offline the normal mode, or the apologetic backup?
When you actually need dictation with no connection
- Long-haul flights. Most aircraft Wi-Fi is spotty at best, and plenty of routes still have none. A transatlantic or transpacific leg is hours of dictation time lost if your tool needs a live connection to work at all.
- Trains and commutes. Tunnels, rural stretches, and dead zones between towers mean the connection drops mid-sentence — repeatedly — even on routes that are "mostly" covered.
- Secured or industrial sites. Government networks, air-gapped systems, hospitals, and factory floors often restrict or block outbound traffic entirely, by policy. A cloud dictation tool simply cannot function there, regardless of signal strength.
- Vans, cabins, and remote work. Rural broadband, satellite links with data caps, or just working somewhere off the grid on purpose — all of these make a connection-dependent tool unreliable at exactly the moment you need it least.
Dictation tools compared: what needs a connection?
| Tool | Where speech is processed | Works with no connection? |
|---|---|---|
| Windows Voice Typing (Win+H) | Microsoft cloud (Azure Speech) | No — required for every sentence |
| Wispr Flow | Cloud (multi-tenant SaaS) | No — required for every sentence |
| Local Whisper apps (e.g. Inkvox) | Your own CPU or GPU | Yes — once the model is downloaded |
The offline-ready checklist
A local dictation tool only helps if it's actually ready before your connection disappears. Run through this before the flight, not during it:
- Download the model on Wi-Fi, in advance. Local speech models are downloaded once and then run entirely from disk — but that first download still needs a connection, so do it the night before, not at the gate.
- Check the model's footprint. Know roughly how much disk space it takes so you're not caught out mid-download on hotel Wi-Fi. Quantized speech models today typically run a few hundred megabytes to under a gigabyte.
- Download the language you actually need. If you dictate in more than one language, confirm which ones are bundled in the base download versus fetched separately — and grab them all before you go.
- Test with the network off. Turn on airplane mode and dictate a sentence before you actually need it to work. If it transcribes normally, you're covered. If anything degrades or refuses, better to find out at your desk than at 35,000 feet.
Where Inkvox fits
Inkvox runs Whisper large-v3-turbo, quantized, directly on your own GPU through Vulkan — NVIDIA, AMD, or Intel — with a CPU fallback if you don't have a compatible card. There's a one-time model download of roughly 800 MB during setup; after that, it works completely offline, in 100+ languages, on a plane, a train, or a network that blocks everything else. Audio is processed in memory on your machine: 0 bytes uploaded, no account required, nothing to reach a server because there's no server in the design — not a fallback, the only mode. That architecture is also why it fits privacy-sensitive work in the first place: offline isn't a feature toggle, it's the whole point. On a mid-range GPU like an RTX 3070, sentences transcribe in roughly 0.3–0.4 seconds — and if you want the deeper hardware picture, see our guide to GPU requirements for local dictation.
The question to ask any dictation tool isn't "does it have an offline mode." It's "what happens the instant the connection disappears" — and for cloud tools, the honest answer is: nothing happens, because nothing works.
Frequently asked questions
Can I dictate on an airplane with no Wi-Fi?
Only with a tool that runs its speech model locally. Windows Voice Typing and cloud dictation apps like Wispr Flow need a live connection for every sentence, so they stop working the moment you lose signal — in airplane mode or otherwise. A local Whisper app with its model already downloaded before takeoff will keep transcribing normally, because nothing needs to reach a server.
Does dictation work without an internet connection at all?
Most mainstream dictation tools do not. Windows' built-in Voice Typing runs on Microsoft's Azure Speech services in the cloud and requires an internet connection by design. Wispr Flow runs entirely in the cloud as well. Dictation without an internet connection requires an app built around a local speech-to-text model — one that processes audio on your own CPU or GPU instead of a remote server.
Is there an offline speech-to-text app for a laptop that doesn't need Wi-Fi?
Yes. Apps built on local Whisper models — the same family of speech-to-text model OpenAI released — run entirely on a laptop's own CPU or GPU. Inkvox is one example: after a one-time ~800 MB model download, it transcribes fully offline on Windows 10/11, on NVIDIA, AMD, or Intel GPUs via Vulkan, with no account and no audio ever uploaded.
How much storage does an offline dictation model need?
It depends on the model, but a modern quantized Whisper model typically needs a few hundred megabytes to just under a gigabyte of disk space. Inkvox's model download is about 800 MB, fetched once during setup — after that, no further downloads are needed to keep dictating offline.
If your dictation tool stops working the moment the Wi-Fi does, join the Inkvox waitlist — one model download, then it runs on your GPU, not someone else's server, wherever you are.