Built-in, local, and cloud dictation are different choices
Windows includes Voice Typing and a separate offline Voice Access feature. Microsoft's current documentation describes Voice Typing's processing in more than one way, and Copilot+ PCs can add on-device Fluid dictation corrections. Dedicated apps add push-to-talk workflows, local model choice, or AI rewriting. This table separates those tradeoffs; it does not rank accuracy or speed because there is no controlled head-to-head benchmark across these products.
| Question | Presspeech | Windows Voice Typing | Windows Voice Access | Handy | Aqua Voice |
|---|---|---|---|---|---|
| Primary workflow | Hold a configurable key, speak, and release to normally paste into the captured window. If that window cannot be verified, published 0.1.12 copies for manual paste; upcoming 0.1.13 instead offers an explicit Delivery Recovery Copy or Discard. Toggle mode is optional. | Press Win+H in a text box; listening starts immediately. Stop with a voice command or the on-screen control. | Keep a voice-control session available to dictate, edit text, and operate Windows with spoken commands. | Hold a shortcut or use toggle mode, then paste the transcript into the active text field. | Hold a key, speak, and release; cloud models refine and insert the result. |
| Where speech is processed | On the PC; the app has no cloud transcription path. Model downloads and optional release checks use the network. Published Windows 0.1.12 may send bundled Hugging Face library telemetry or a locally available token with a model request; see the version-specific network inventory. | Microsoft's Voice Typing support page says it uses online Azure Speech; Microsoft's privacy statement says Windows 11 Voice Typing may use both device-based and online recognition. Copilot+ PCs can add on-device Fluid dictation corrections after its model is installed. That correction feature alone does not establish that all speech recognition is offline; check the current Windows settings and documentation for your device. | On the PC after its speech language files are installed; Microsoft documents that Voice Access works without an internet connection. | Core transcription runs on the computer. Models download during setup; optional post-processing can use a configured local or cloud provider. | In the cloud, with no offline mode. Aqua's Privacy Mode controls transcript retention but does not move recognition onto the PC. |
| Text handling | Deterministic dictionary replacements, voice shortcuts, filler removal, and British spelling. No AI rewriting. | Automatic punctuation plus spoken punctuation, editing, and keyboard commands supported by the selected language. When enabled on a supported Copilot+ PC, Fluid dictation uses an installed on-device model to correct grammar, punctuation, spelling, and filler words (English locales). | Dictation, selection, correction, formatting, punctuation, and PC-control commands in one accessibility workflow. | Dictionary and model controls, transcript history, and optional post-processing alongside local transcription. | AI punctuation and formatting, filler cleanup, a custom dictionary, and writing-style instructions. |
| Languages | Model-dependent. CUDA Parakeet is multilingual; a fresh CPU-only installation starts with English-only Whisper base.en, with other local models selectable. | A broad Microsoft-published language and locale list; switch by changing the Windows input language. | Microsoft currently lists English, Spanish, French, German, Chinese, and Japanese language families. | Model-dependent local transcription, including multilingual Whisper and Parakeet choices. | 49 languages according to Aqua's Windows page. |
| Price and source | Free and MIT licensed. | Included with Windows; proprietary. | Included with Windows 11; proprietary. | Free and MIT licensed. | 1,000 words free; Pro is $8/month billed annually. Proprietary. |
| Setup and scope | Windows 11 x64 is recommended. Windows 10 x64 remains compatible only while its edition is supported or enrolled in Extended Security Updates; general support ended in 2025. The self-contained prerelease is about 4.4 GB plus its model and is currently unsigned. | Built into Windows 10 and 11. It needs a microphone, internet access, and cursor focus in a text box. | Built into Windows 11 22H2 and later. First setup downloads a speech model; it is designed for full voice access, not only push-to-talk dictation. | A cross-platform desktop app with selectable engines and models. Packaging and acceleration vary by hardware and operating system. | A Windows 10/11 service tied to an Aqua account and subscription allowance; one subscription also covers its Mac and iPhone apps. |
Start with Voice Typing if zero installation matters, but check the processing behavior available on your Windows version and device before dictating sensitive material. Its optional Copilot+ Fluid dictation feature applies on-device corrections; do not assume this means the entire voice-typing recognition path is offline. Try Voice Access if you need documented offline dictation as part of full hands-free Windows control and it supports your language. Compare Presspeech and Handy when local push-to-talk and open source matter: Presspeech keeps a fixed per-platform path and no transcript library, while Handy prioritises model choice, one cross-platform app, and history. Aqua Voice is the different trade: cloud processing and a subscription in exchange for AI refinement and a shared commercial service across devices.
Another native open-source option: Scribe AI is the closest Windows-only peer: it bundles local multilingual Parakeet, supports x64 and Arm, and has a Microsoft-signed Store install. It also deliberately offers a broader product surface than Presspeech, including persistent local history, per-app profiles, usage views, and optional local or user-configured cloud AI cleanup (Scribe sources checked 22 September 2026).
Another cross-platform local option: Superscribe Lite offers Mac and Windows downloads, local recognition without an account, live text insertion, and optional text cleanup through a provider key supplied by the user. Its features and privacy boundary are the vendor's own claims; the focused comparison separates those from Presspeech's after-recording workflow and version-specific model-download notices (Superscribe sources checked 23 September 2026).
Sources: Presspeech Windows guide and source; Microsoft documentation for Voice Typing usage, languages, and Copilot+ Fluid dictation, speech processing and privacy, the Windows privacy statement, and Voice Access scope, requirements, and languages (checked 23 September 2026); Handy source and documentation; Aqua Voice's Windows, pricing, processing, and language page and privacy policy (checked 1 September 2026); Scribe AI's source and documentation, privacy policy, and Microsoft Store listing (checked 22 September 2026); and Superscribe Lite's official launch note (checked 23 September 2026). Vendor accuracy claims are deliberately omitted because the products do not publish a controlled benchmark against one another.