Plain-language explainers for the technology behind desktop dictation — from speech models and text injection to privacy and per-language accuracy.
Inside the cleanup pass that defines 2026-era dictation: filler removal, punctuation, formatting and tone adjustment by a language model after transcription — why it feels like magic and where it goes wrong.
The OS-level story of text injection: macOS Accessibility APIs versus synthesised keystrokes versus the pasteboard trick, Windows UI Automation and SendInput, and why Wayland broke the Linux version of this.
How dictation apps listen for a hotkey in every app at once, why macOS makes them ask for three separate permissions, and why an OS update can silently break everything.
The complete journey of a dictated sentence: capture, voice activity detection, chunking, the encoder-decoder model, streaming partials, the cleanup pass, and text injection.
What actually determines dictation accuracy before any AI is involved: microphone choice and placement, room noise, speaking rate, and the voice-activity detection that decides when you're talking.
The 2026 engine landscape: Whisper's 99-language legacy, Parakeet V3's European sprint, and the proprietary cloud models pulling ahead at the top end — plus how to find out which engine your app actually uses.
The state of offline dictation in 2026: local Whisper and Parakeet models, RAM and CPU requirements by model size, the real quality gap versus cloud, and which apps are genuinely offline versus merely offline-ish.
What dictation apps actually do with your audio and transcripts: cloud retention policies, training opt-outs, what 'on-device' does and doesn't guarantee, and how to choose for clinical, legal or confidential work.
How language support really works in dictation apps: why so many apps share the same 99-language list, what Parakeet V3 changed for European languages, and how to tell real support from a marketing count.
Which dictation apps handle Spanish properly: inverted punctuation, regional accents, tuteo vs usted register, and Spanish-English code-switching, tested across the full ranking.
A transparent look at DictationRank's ranking model: the hard gates (platform and privacy stance), the weighted scores across five use cases and per-language accuracy, and a worked example.
The popularity league table for desktop dictation, built from the only public signals that exist: App Store rating counts, GitHub stars, and vendor-published user numbers.
The state of voice coding in 2026: why identifiers and symbols break general dictation apps, how command-mode and context-aware apps differ, and practical setups for wrists that need a break.