Direct answer — Is Whisperstream a good offline dictation app for Windows? Yes, if you only use Windows. Whisperstream transcribes speech on your CPU with NVIDIA Parakeet, never uploads your audio, and costs $29 once for one PC or $49 for two. Its optional AI cleanup can run locally or through a cloud provider you configure, so check Lockdown Mode if nothing may leave the machine. Mac users need another tool.
Most dictation apps now bill monthly and send your voice to a server. This Whisperstream review looks at a Windows app that does neither: you pay once, and speech recognition stays on your PC. We cover what Whisperstream offline mode really means, what the Whisperstream pricing buys you, where the few network connections go, and how it compares with Weesper Neon Flow. If you have been hunting for offline dictation on Windows at a one-time price, it deserves a spot on your shortlist — with a couple of caveats worth knowing before you install it.
What is Whisperstream, and who is it for?
Whisperstream is a push-to-talk dictation app for Windows 10 and 11 (64-bit) that transcribes on your own computer. It is built by Lanreal Technologies Inc., a company based in Toronto, Canada.
Offline dictation here means that the speech-to-text step runs on your device, with no audio sent to a cloud transcription service. The text is then pasted or typed into whichever application has focus.
The product is aimed at people who want three things at once:
- A Windows-native tool, with no macOS or Linux build
- Local speech recognition, with no account tied to your voice
- A one-time licence instead of a subscription
According to its changelog, version 0.1 shipped on 15 April 2026 and version 1.5.1 on 6 September 2026. That is a young product with a fast release pace.
How does Whisperstream dictation work on Windows?
You hold a hotkey, speak, and release; the transcript lands at your cursor. The default key is Right Shift, and you can remap it or switch to a toggle mode for hands-free dictation.
Text arrives by clipboard paste, or by simulated keystrokes for apps that block pasting. The vendor lists Outlook, Word, Slack, VS Code, browsers and terminals as working without an add-in. It also notes that security, administrator and remote-session policies can still block delivery, so test it on locked-down machines first.
The main features, as published on the vendor’s site:
- Custom dictionary for names, jargon and acronyms
- Per-app profiles, each with its own AI prompt, switched automatically by application or website
- Encrypted local transcript history, with search and audio replay
- Audio file import, transcribed on your PC
- 39 languages, from English and French to Japanese, Arabic and Vietnamese
- Lockdown Mode for high-sensitivity work
One detail matters if you work across remote sessions: the keystroke input mode, added in version 1.0.1, is described as working in Citrix, RDP and screen-reader workflows when policies allow simulated input. If you are still weighing Windows’ own tools, our Windows 11 voice dictation setup guide covers the built-in options first.
Which speech model does Whisperstream use?
It uses NVIDIA’s open-weight Parakeet model on the CPU, plus Qwen3 ASR for languages Parakeet does not cover. Despite the name, it is not built on OpenAI’s Whisper.
The vendor states that Parakeet covers 25 languages and Qwen3 ASR adds 14 more, which is how it reaches 39. Since version 1.5, you can pick between the two when both support your language.
Whisperstream does not name the exact Parakeet checkpoint. For reference, NVIDIA’s current multilingual model card, parakeet-tdt-0.6b-v3, describes a 600-million-parameter model covering 25 European languages under a CC-BY-4.0 licence. For a deeper look at how the two model families differ, read our Parakeet vs Whisper comparison.
What hardware do you need?
A reasonably recent Windows PC, and no graphics card. The published requirements are:
| Requirement | Whisperstream (vendor figures) |
|---|---|
| Operating system | Windows 10 or 11, 64-bit |
| RAM | 8 GB minimum, 16 GB recommended |
| Processor | Runs on CPU; Intel 8th gen+ or AMD Ryzen 3000+ recommended for near-instant results |
| GPU | Not required; used only by the optional local AI cleanup |
| Account | Not required |
Is Whisperstream really offline?
Speech recognition is; the app as a whole is not fully air-gapped. The vendor’s own wording is precise: “Core transcription makes no network call and does not upload your audio.”
The same page lists the connections that remain:
- Licence activation and validation
- Software updates and model downloads
- Voluntary feedback and reviews
- The Agent Speak companion for Claude Code and Codex
- Any online cleanup provider you configure
The privacy policy adds useful detail. Licence validation goes to Polar, the merchant of record, about once every seven days while you are online, and sends your licence key and your device’s hostname. That is narrow operational metadata, not voice data, but it is worth knowing if your hostnames are sensitive.
In short, your audio stays on your PC. Dictation keeps working on a plane or when the Wi-Fi drops, subject to your licence state. For the wider privacy argument behind this design, see why offline dictation protects your privacy.
Does the AI cleanup send your text to the cloud?
Only if you set it up that way. AI cleanup is optional and off by default, and you choose where it runs.
When enabled, it removes filler words, fixes grammar and punctuation, and applies spoken corrections — say “Friday, sorry Saturday” and only “Saturday” survives. The features page lists three possible engines:
- A downloadable local model, on supported hardware with a capable GPU
- Ollama, running on your own machine
- A cloud or OpenAI-compatible provider that you configure
In the cloud case, the provider receives transcript text, not audio. The policy also notes that imported audio files stay local, but their transcripts can enter cloud cleanup if you enabled it.
The practical advice: if your work is confidential, turn on Lockdown Mode. According to the vendor, it limits AI cleanup to an on-device model, switches off Whisperstream’s cloud enhancement, and can put your history behind a password. Administrators can enforce it across restarts, updates and reinstalls.
What does Whisperstream cost?
$29 once for one device, or $49 once for up to two of your own devices. There is no subscription and no recurring licence fee.
| Whisperstream | Weesper Neon Flow | |
|---|---|---|
| One-time price | $29 (1 device) or $49 (2 devices) | €99 lifetime |
| Recurring options | None | €5/month or €45/year |
| Devices per licence | 1 or 2 | Up to 3 of your own computers |
| Free trial | 7 days, no account or card | 15 days, no credit card |
| Refund window | 30-day money-back guarantee | 7-day money-back guarantee |
| Platforms | Windows 10/11 (64-bit) | macOS and Windows |
The trial length comes from the vendor’s changelog and privacy policy, which both describe a seven-day trial; the home page itself just says “free trial”. Prices are shown in the currency each vendor publishes, so the two columns are not directly comparable without conversion.
On pure entry price for a single Windows PC, Whisperstream is hard to beat. If you are deciding between paying once and paying monthly in general, our guide to subscription vs one-time purchase dictation apps sets out the trade-offs. You can see every Weesper plan on the pricing page.
Is the installer trustworthy?
The installer is Authenticode-signed, under the publisher name Lanreal Technologies Inc. Authenticode is Microsoft’s code-signing scheme: it identifies the publisher and confirms the file has not been altered since it was signed.
That matters for IT teams, who can allowlist the exact publisher name on managed machines. The vendor openly notes that SmartScreen may still warn while a new binary builds reputation, which is normal for a young product.
Whisperstream vs Weesper: how do they compare?
Both keep speech recognition on your computer, so the Whisperstream vs Weesper question comes down to platforms, model, languages and billing.
| Whisperstream | Weesper Neon Flow | |
|---|---|---|
| Platforms | Windows only | macOS and Windows |
| Speech model | NVIDIA Parakeet + Qwen3 ASR (CPU) | Whisper, running locally |
| Windows acceleration | CPU only for transcription | CPU, or Vulkan GPU on compatible PCs |
| Languages | 39 | 55+ |
| Audio leaves the device | No | No |
| Custom vocabulary | ✅ | ✅ |
| Per-app AI prompts | ✅ | ✅ Custom prompts |
| Billing | One-time only | Monthly, yearly or one-time |
Choose Whisperstream if you work only on Windows, dictate in one of its 39 languages, and want the lowest one-time price for one or two PCs.
Choose Weesper Neon Flow if you switch between a Mac and a Windows PC, need a language outside Whisperstream’s list, or want one licence to cover up to three of your own computers. Weesper runs Whisper locally on both systems, and on Windows it can use a compatible Vulkan GPU for faster transcription.
Whisperstream is not the only one-time offline option either. CamoVoice also sells a perpetual licence and runs on Mac and Windows, with a smaller language list.
Want to compare on your own machine? Download Weesper Neon Flow and run both side by side during your trials.
What are Whisperstream’s limitations?
The main one is platform: it is Windows only, with no macOS, Linux or mobile version.
The others are smaller, but worth weighing:
- Short trial. Seven days is tight for judging accuracy across real meetings and documents.
- Young product. The first release dates from April 2026, so long-term support is still unproven.
- Local cleanup needs a GPU. Without a capable graphics card, AI cleanup means Ollama or a cloud provider.
- Nuanced offline story. Licence checks and updates still need the internet, as with most paid desktop apps.
- Two devices maximum. The largest personal licence covers two PCs; the vendor handles teams through corporate licensing.
None of these is a deal-breaker for a Windows user who wants private, pay-once dictation. They simply define who the product is for.
Final verdict: should you buy Whisperstream?
Whisperstream is one of the more transparent offline dictation apps for Windows. Speech recognition runs on your CPU, your audio is not uploaded, the network connections that remain are documented, and the price is $29 once. Lockdown Mode gives sensitive users a clear switch to keep AI cleanup on the device.
If Windows is your only platform, it is a sensible pick. If you work across a Mac and a PC, or need more languages and a longer trial, Weesper Neon Flow covers both systems with the same local Whisper engine.
Ready to try private dictation on your own terms? Try Weesper free for 15 days on macOS or Windows, and compare it with your current setup before you commit.