Most "best dictation app" roundups rank on words-per-minute and accuracy, then mention privacy in a footnote. That gets the priority backwards. If you dictate client email, legal notes, patient summaries, unreleased code, or anything you signed an NDA over, the most important spec isn't speed — it's whether your voice ever leaves your computer.
And here's the uncomfortable part: most of the popular dictation tools on Windows send your audio to the cloud by default. Some discard it immediately. Some keep it. A few quietly use it to train models. "Private" has become a marketing word stretched over very different architectures, and the difference matters precisely when you can least afford to be wrong about it.
This list ranks Windows dictation tools by the only question that actually defines privacy: where does the audio go? We'll be honest about the tradeoffs — the most private options often ask more of you, and the slickest options often ask for your data.
Why "private" is the real buying criterion
There are two fundamentally different ways a dictation app can turn your speech into text.
On-device (local) transcription runs the speech model on your own machine. Your audio is captured, converted to text locally, and never transmitted. There's no server to subpoena, no retention policy to trust, no breach that can leak what you said. This is the only architecture where "private" is a property of the system, not a promise on a settings page.
Cloud transcription uploads your audio (or a stream of it) to a remote server, where a model transcribes it and sends text back. This is usually faster and more accurate, because the server has more compute than your laptop. But now your words have traveled across the internet to a company's infrastructure, governed by that company's retention policy, subprocessor list, and terms of service — all of which can change.
Most cloud tools will tell you they encrypt data in transit and at rest, and that's good and true. But encryption protects data from outsiders; it does nothing about the vendor itself having your audio and deciding what to do with it. "Zero data retention" is a meaningfully stronger promise — but it's still a policy, not a guarantee baked into where the computation happens. A policy can be revised. Physics can't: if the audio never left your machine, there's nothing to retain.
That's the lens for this list. We're not anti-cloud — cloud transcription is genuinely excellent — we're just refusing to call something "private" when the defining fact is that your voice goes somewhere you don't control.
The criteria we judged on
Every tool below was scored on five things:
- Local / offline option. Can transcription happen entirely on-device, with no audio leaving the machine? This is weighted heaviest.
- Data retention and training. If it does use the cloud, is there a real zero-retention mode, and is your voice excluded from model training?
- Works in every app. Does it inject text system-wide — into Outlook, Slack, your browser, your IDE — or is it trapped inside one editor or one website?
- Action features beyond typing. Can it do more than drop text in a box: send a message, translate on the fly, query an AI, run a command? Dictation is table stakes in 2026; what you do with the words is where these tools separate.
- Windows-native and easy to start. It has to actually ship on Windows (which rules out some excellent Mac-only tools), and the friction to get going matters.
A quick note before the list: two of the most-recommended "private Whisper" apps, superwhisper and MacWhisper, are genuinely local and genuinely good — but they're macOS-only, with no Windows version planned. (whisperdictator.com) If you're reading this on a Mac, look there for now. (davr is Windows-only today, with Mac, iPhone, and Android rolling out over the next ~60 days — but it isn't available on those platforms yet, so we're not counting it for Mac here.) For Windows, here's the field.
The ranking
1. davr — local transcription plus an action layer
Verdict: The strongest fit if you want on-device privacy and want dictation to actually do something. davr is Windows-first, with a local Whisper option that runs transcription fully offline — the audio never leaves your machine. (Source: davr product facts) When you do want the convenience of the cloud, it uses OpenAI Whisper for speech and Anthropic's Claude to clean up filler words and format what you said.
What sets it apart isn't only the privacy mode — it's the action layer. davr injects text into whatever app is focused, but it also translates as you dictate (speak one language, output another, across ~40 dictation languages), triggers an "ask Claude" query from a hotkey to drop an answer into your active window, and includes Veil — private steganographic messaging that buries a secret note inside innocent-looking cover text only your contact can decode (AES-256 + X25519, with a fully offline Veil Local mode). (Source: davr product facts) So instead of "dictate, then go do the thing," the dictation is the thing.
Privacy note: With the local option on, audio is transcribed on-device and isn't transmitted. API keys (for the cloud path) are stored in Windows Credential Manager rather than in plaintext. (Source: davr product facts)
Why it's #1 here, honestly: not because it's the fastest or most accurate — we're not going to quote an accuracy number we can't independently cite. It earns the top slot on this list because it's one of the few Windows tools that combines a true on-device option with system-wide injection and an action layer, rather than forcing you to choose between "private" and "useful." If your only need is raw transcription speed and you don't care where audio goes, a cloud tool may suit you better. On pricing: a Free tier costs $0 if you bring your own OpenAI/Anthropic key (dictation only), and paid plans start at Pro at $6/mo bring-your-own-key or $12/mo managed ($66/yr and $132/yr annually), with a Max tier at $10/mo BYOK or $16/mo managed. Every paid plan can also be tried free for 14 days with no credit card. (Source: davr product facts)
2. Talon Voice — free, fully local, and not for the impatient
Verdict: The most powerful free local option on Windows, if you're willing to climb the learning curve. Talon runs a local speech engine (its Conformer model) entirely on-device, no internet required. (talonvoice.com) It was built for hands-free computer control and coding-by-voice — people with RSI, accessibility needs, and developers who drive their whole machine with their voice swear by it.
Privacy note: Local-first by design; audio is transcribed on-device. (spokenly.app)
The catch: Talon is a command-and-control system, not a drop-in dictation box. Expect to learn a grammar and tinker with config. It's free to use, with an optional paid tier for extra features. (spokenly.app) If you want to configure your voice setup, it's brilliant. If you want to install something and start dictating in five minutes, it isn't that.
3. Dragon Professional v16 — the expensive incumbent that stays on-device
Verdict: The original serious dictation engine, and still a legitimate private choice for regulated professions. Dragon Professional v16 processes speech mostly on-device after you set up a voice profile, works offline, and ships deep domain vocabularies for legal and medical work. (shop.nuance.com)
Privacy note: On-device processing means your dictation doesn't need to traverse the cloud — a big reason it persists in law firms and clinics. (Note: Dragon Anywhere and Medical One are cloud products; Professional is the local one.)
The catch: It's a roughly $699.99 one-time license (getvoibe.com), the interface shows its age, and it's a single-purpose desktop tool — no translation, no AI command layer, no modern messaging actions. You're buying a vault, not a Swiss Army knife.
4. Local Whisper, DIY — Whisper.cpp, Buzz, and friends
Verdict: Maximum privacy, minimum hand-holding. Open-source projects like Whisper.cpp, Buzz, and nerd-dictation run OpenAI's Whisper models locally on Windows, free, with nothing sent to the cloud. (clevertype.co)
Privacy note: Fully local — the audio never leaves your machine, and you can verify it at the network level.
The catch: There's no company behind it, no support line, and system-wide text injection usually isn't turnkey — you're assembling a workflow, not installing a product. Great if you're technical and want total control and zero cost. Frustrating if you just want to talk into your email.
5. Windows 11 Voice Typing — free, built-in, private only if you flip a switch
Verdict: Already on your PC, and it can be private — but you have to know how. By default, Voice Typing uses Online Speech Recognition, which streams your audio to Microsoft's Azure cloud. (support.microsoft.com) Turn Online Speech Recognition off (Settings → Privacy & security → Speech) and Windows falls back to device-based recognition with no voice data sent to Microsoft. (support.microsoft.com)
Privacy note: Device-only mode is genuinely local; the default cloud mode is not. Most people never change the setting, so most people are dictating to Azure without realizing it.
The catch: Offline accuracy is noticeably weaker, there's no formatting cleanup, and there are no action features. It's a competent free baseline, not a destination.
6. Wispr Flow — fast, polished, and cloud, always
Verdict: The slick, popular option — and the clearest illustration of why "private" needs a definition. Wispr Flow is genuinely good: low-latency, accurate, works across your apps, and available on Windows. But its transcription always happens in the cloud — every dictation uploads audio to AWS, processed through subprocessors including OpenAI, Anthropic, and others. (wisprflow.ai/privacy, docs.wisprflow.ai)
Privacy note — read this carefully: Wispr offers a Privacy Mode that enables zero data retention — audio is transcribed and discarded, not stored, and not used for training. (wisprflow.ai/privacy) That's a strong, commendable policy. But it is not the same as on-device. Your audio still leaves your machine and is processed on someone else's servers; Privacy Mode governs what happens after it arrives, not whether it travels. For many people that's a fine trade. For anyone who needs the audio to never leave the building, it doesn't clear the bar.
The catch: It's a subscription — $15/mo, or $12/mo billed annually, with a free tier capped at 2,000 words/week and a 14-day trial. (wisprflow.ai/pricing) If cloud transcription is acceptable to you, Wispr is a strong pick. If it isn't, no policy setting changes the architecture.
7. Google Docs Voice Typing — convenient, and entirely in Google's cloud
Verdict: Free and right there in Docs, but it's the opposite of private and the opposite of system-wide. Voice Typing streams every word you speak to Google's servers for recognition, with no offline mode, and it only works inside Google Docs (primarily in Chrome). (support.google.com)
Privacy note: Cloud-only. Google sees everything you dictate while it's on. Fine for a grocery list; wrong for anything confidential.
The catch: Locked to one app and one browser, no formatting intelligence, no actions. Convenience tool, not a privacy tool.
8. Otter.ai — a meeting recorder, not a dictation box
Verdict: Worth naming so you can rule it out for this job. Otter is built for transcribing meetings, not for dictating into your apps; audio uploads to its cloud, and the company has faced a 2025 class-action complaint (Brewer v. Otter.ai) alleging it recorded conversations and used data for AI training without proper consent. (otter.ai, workplaceprivacyreport.com)
Privacy note: Cloud-first, and the litigation is exactly the kind of risk a privacy-conscious buyer is trying to avoid.
The catch: It's not a system-wide dictation app at all. Different category — included here only because it surfaces in dictation searches and shouldn't be your answer.
How to choose
Strip away the noise and it comes down to one fork:
If your audio must never leave your PC — you handle confidential, regulated, or NDA-bound text — you want a tool with a true on-device option. That's davr (if you also want system-wide injection and action features), Talon (if you're technical and want it free), Dragon Professional (if you're in law/medicine and have the budget), or DIY local Whisper (if you'll assemble it yourself). Windows Voice Typing with Online Speech Recognition turned off is the free floor.
If cloud transcription is acceptable and you mainly want speed and polish, Wispr Flow is the strongest of the cloud-first tools, especially with Privacy Mode on — just don't mistake zero-retention for on-device. Google and Otter are convenient for their narrow jobs but aren't built for private, system-wide dictation.
The trap to avoid is treating "encrypted" or "zero retention" as if it meant "local." They're better than nothing, but they're promises about a server that still has your voice. If that distinction matters to you, the shortlist is short — and that's exactly the point of this list.
Try davr
davr is the rare Windows tool that doesn't make you choose between private and useful: a local transcription option so your audio can stay on your machine, plus an action layer that turns your voice into translations, AI queries, and encrypted Veil messages across every app you use. Connect your own OpenAI/Anthropic key and dictation is free — $0. Want the managed AI features without wiring up keys? Try them free for 14 days, no credit card. Either way, see whether dictation that keeps your words on-device, and then actually does something with them, changes how you work.