Start with the platform you use. The macOS table compares five tools on the same questions; the Windows guide separates Microsoft's online and offline built-ins from dedicated local and cloud apps. Product claims come from published sources, and Presspeech performance claims come from its reproducible macOS benchmark.
| Category | Presspeech | Apple Dictation | Superwhisper | Wispr Flow | VoiceInk |
|---|---|---|---|---|---|
| How it works | Hold a key, speak, release; the verbatim transcript pastes at the cursor. | Press the Dictation shortcut to start, stop manually; auto-stops after 30 s of silence. | Hotkey recording; optional AI modes reshape the output before insertion. | Hold hotkey; AI-formatted text appears. | Hotkey dictation with optional AI enhancement. |
| Price | Free. | Included with macOS. | Free tier; Pro starts at $8.49/month. | Free tier of 2,000 words/week on desktop; Pro is $15/month, or $12/month billed annually. | From $25 one-time for one Mac; free to build from source. |
| Source | Open source, MIT. | Proprietary, built into macOS. | Proprietary. | Proprietary. | Open source, GPL-3.0 per its README. |
| Where audio is processed | On this Mac only; no cloud transcription path exists in the app. | On-device for supported languages; otherwise sent to Apple servers. Keyboard settings shows which applies. | Your choice per model: fully local models or cloud services. | In the cloud — “Transcription always happens in the cloud” per its privacy page. | On-device by default; optional cloud providers. |
| Account required | No. | No. | No sign-in documented; Pro activates with a license key. | Yes. | No; the paid binary activates with a license key. |
| Published latency evidence | Model-only p50 of 92–152 ms across the published clips; the reproducible benchmark does not time clipboard or paste work. | Not published. | Not published. | Its engineering blog describes a ~700 ms end-to-end target, which is not directly comparable with Presspeech's model-only measurement. | Not published. |
| App size | 8.4 MB release zip, plus about 500-600 MB for the local model cache. | Built in; on-device languages need a model download, size not published. | Not published; local models are 75 MB–3 GB. | Not published; docs ask for ~500 MB free storage. | 47.8 MiB download (v2.13 DMG); models extra. |
| Idle footprint | ~80 MB RAM, 0% CPU between dictations. | Not published. | Not published. | Not published; 8 GB+ RAM recommended. | Not published; 8 GB RAM recommended. |
| Languages | 25 European languages, auto-detected; Settings can pin one of 18. | 63 Dictation locales, 48 of them with on-device processing. | 100+ via cloud models; local model counts vary. | 100+. | No official count; its Whisper models cover about 100. |
| Platforms | Released on Apple Silicon Macs, macOS 14+; a separate Windows preview is available. | Every Mac. | macOS 13.3+, Windows, iOS; local models work best on Apple Silicon. | macOS 11+ including Intel, Windows, iPhone, Android. | Apple Silicon Macs, macOS 14.4+. |
| Beyond verbatim dictation | Deterministic local corrections and shortcuts, opt-in spoken formatting, and filler-word removal; no AI rewriting. | Auto-punctuation, spoken punctuation and emoji commands. | AI modes, meeting transcription, file transcription, translation. | AI formatting, tone styles, command mode, cross-device dictionary. | AI enhancement modes, per-app Power Mode profiles, file transcription. |
Every tool here wins somewhere. Pick the AI-formatting tools if you want your words reshaped, Apple Dictation if you want zero install, VoiceInk if you want one configurable local Mac app, or Handy if you want a larger cross-platform open-source community and model choice. Presspeech's case is narrow on purpose: a measured low-latency local model path, a small native macOS app, and no cloud transcription path to audit.
Presspeech's figures are measured on an M4 Mac mini and reproducible from the benchmark harness. Other tools' figures come from their official sites, docs, and release assets, checked 1 September 2026. “Not published” means we could not find an official number, nothing more. MacWhisper sits out this table because it is primarily a file-transcription app; it has its own page. Handy is the closest high-adoption open-source alternative and has its own page rather than making this Mac-focused table wider. Spotted an error? Open an issue and we will fix it.
Sources: Presspeech source and benchmarks; Apple Dictation guide, macOS feature availability, and Siri & Dictation privacy; Superwhisper site and docs; Wispr Flow pricing and privacy; VoiceInk site and v2.13 release; Handy source and documentation.
Compare Presspeech with Windows Voice Typing, offline Voice Access, Handy, and Aqua Voice. The guide distinguishes processing location, trigger workflow, cost, language scope, and installation tradeoffs without claiming an unmeasured accuracy winner.
Both are free, MIT-licensed, and local on Windows. Handy offers one cross-platform application and broader model choice; Presspeech uses a separate Windows pipeline with deterministic text handling.
Single-purpose free pipeline on the Neural Engine versus a configurable dictation workspace with paid tiers.
Free on-device push-to-talk versus a cloud-transcription subscription with AI formatting.
Two open-source local dictation apps: a fixed minimal pipeline versus a configurable multi-model workspace.
Two free MIT-licensed local dictation apps: native fixed pipelines versus a cross-platform multi-model application.
Small focused third-party menu-bar workflow versus the built-in macOS dictation feature.
Push-to-talk dictation utility versus a broader transcription app.
End-user Parakeet/ANE utility versus apps and projects built around Whisper-family models.