Hebrish
Experimental Early days, and every version gets better. Roadmap

Ubuntu app

Screenshots

What Hebrish looks like on Ubuntu: version 1.011, captured 2026-10-29. Click a screenshot to see it larger, then use the arrow keys (or swipe) to go through them all.

Voice Typing and Transcription

Hebrish 1.011 on Ubuntu, Voice Typing tab: a round Talk button, the microphone with its level meter and Test mic, voice detection switched on, a text box, and the output choices At the cursor, Clipboard and This window.
Voice TypingPress your toggle key in any app, or tap Talk, and speak: the text goes where your cursor is. Output can go at the cursor, to the clipboard and into this window, in any combination. Voice detection skips silent takes and trims long pauses.
Hebrish 1.011 on Ubuntu, Transcription tab: Record, Pause, Retake, Stop and Transcribe buttons, voice detection and save-audio options, Open an audio file, and an empty transcript box with Paragraphs set to Paragraph model.
TranscriptionFor longer sessions: record with Pause and Retake, or open an audio file (or drop one on the tab). The transcript lands in a box you can edit, split into paragraphs by the paragraph model if you choose; copy it or save it.

Models

Hebrish 1.011 on Ubuntu, Models tab: the Sabich, Jachnun, Shakshuka and Parakeet (stock) families; Sabich v2.2 in use and recommended, 630 MB, and Sabich-Parakeet-110M below it, about 137 MB, each with speed and accuracy dots.
ModelsPick a family (Sabich, Jachnun, Shakshuka or stock Parakeet), then a version, by its speed and accuracy dots. Sabich is the recommended one. A model downloads once and then works offline. Shown: Sabich v2.2 in use, and Sabich-Parakeet-110M, a fifth of the size, for older phones and wearables.
Hebrish 1.011 on Ubuntu, Models tab, Utilities: Silero VAD for voice detection, included with the app, and the Paragraph model for splitting long transcripts, downloaded.
UtilitiesSmall models that work alongside the speech model, offline. Silero VAD, included with the app, tells speech from fans, traffic and café noise. The paragraph model splits long transcripts where the topic moves on.

History and Dictionary

Hebrish 1.011 on Ubuntu, History tab: word counts for today, the last 7 days and all time, a search box, recent transcripts with copy, edit and delete buttons, and where the history is stored.
HistoryEvery transcript, with word counts for today, the last 7 days and all time. Search it (beginnings match: “shak” finds Shakshuka), then copy, edit or delete a take, or export the lot. Text only, no audio, and it stays on your computer.
Hebrish 1.011 on Ubuntu, Dictionary tab: fields to add a word and what the model hears instead, and an empty list of your words.
DictionaryWords you want written your way: names, brands, Hebrew terms. Add what the model writes instead (“Heard as”) and Hebrish corrects it from the next take. It fixes the text after the model; it can’t teach the model new sounds.

Settings

Hebrish 1.011 on Ubuntu, Settings, Voice Typing: the toggle key set to Wakeup, the push-to-talk key, double-tap to retake, Esc to discard, the paste shortcut, and output options.
Hotkeys and outputA toggle key and a push-to-talk key: any key on any keyboard or foot pedal (F13–F24 are ideal). Double-tap to start again, Esc to discard, and options such as a space after each transcript or ALL CAPS.
Hebrish 1.011 on Ubuntu, Settings, Microphones: every microphone the app has seen, with a name, type, how it was detected and whether it is connected, in a preferred order.
MicrophonesEvery microphone Hebrish has seen on this computer, with a name and a type, in the order you want them used: each take uses the first one that is connected.
Hebrish 1.011 on Ubuntu, Settings, Action keys: optional keys for Start, Stop, Pause / resume, Retake, Discard, Stop and send, and Paste last transcription, plus the send key.
Action keysOptional keys for a macropad or foot pedals: Start, Stop, Pause, Retake, Discard, Stop and send, and Paste last transcription. Stop and send waits for the last words, then presses the send key, so nothing is cut off in a chat box or a terminal.
Hebrish 1.011 on Ubuntu, Settings, Sounds and Floating overlay: beeps or a voice for start and stop, volume, quiet hours, and the overlay’s show and opacity options.
Sounds and the floating overlayBeeps or a spoken voice for start and stop, with volume and quiet hours. The floating overlay is a small bar over every app with the timer, a level meter, Pause, Retake and Discard; it never takes the keyboard focus.
Hebrish 1.011 on Ubuntu, Settings, Model updates, App updates and System: check every 24 hours, install new model versions automatically, check for a new app version, and start at login in the tray.
UpdatesNew model versions install automatically, and the old version is deleted once the new one has downloaded. The app checks for a new version of itself; updates come from the Hebrish repository, or Software Updater like any other package.

Get Hebrish for Ubuntu

Every screenshot is the real app, unedited, on Ubuntu with KDE. Screenshots are refreshed with app releases; this set is version 1.011.