← statichum.studio

Local, Private Dictation With Per-Speaker Training And Accent Robustness, The One Dragon NaturallySpeaking Feature Nobody Rebuilt For Whisper

desktop app real project • single request

People with arthritis, RSI, or limited hand dexterity now have plenty of local Whisper-based dictation tools on Linux, but every one of them dropped the feature that made Dragon NaturallySpeaking usable for this audience: per-speaker adaptive training, plus solid handling of strong or non-US accents. The accessibility wedge isn't raw model accuracy, it's a fine-tune-on-your-own-voice flow that runs locally and works system-wide. Underserved disabled and heavy-accent users are explicitly asking for it.

builder note

Everyone rebuilt Whisper dictation and skipped the feature that mattered for this audience: per-speaker adaptation. A fine-tune-on-your-voice flow for accents and atypical speech is the wedge, not another point of accuracy.

landscape (4 existing solutions)

Local Whisper dictation is now plentiful, but every option skipped the speaker-adaptive training that made Dragon usable for people with heavy accents or atypical speech, which is exactly the accessibility audience asking here.

Speech Note (Linux Flatpak) Strong local accuracy (~95% without GPU) but no per-speaker adaptive training and shaky on non-US or atypical accents.
Speed of Sound / Handy Global-hotkey dictation that inserts text anywhere on the Linux desktop, but uses one-size-fits-all Whisper/Parakeet models with zero voice personalization.
FUTO Voice Excellent local dictation on Android, but no desktop/global Linux integration and no speaker training.
Dragon NaturallySpeaking The only tool that ever did deep speaker adaptation is Windows-only and effectively abandoned for the Linux/accessibility crowd.

sources (1)

other https://lemmy.world/post/47731678 "I'd love something that can be trained by my speech like Dragon" 2026-06-04
accessibilityspeech-to-textlocal-firstlinuxprivacy