Linux voice input

voxi types what you say, as you say it.

High-quality speech-to-text dictation for Linux and Wayland, with continuous eager sentence streaming, physical modifier-key safety and direct typing into the active window. Audio stays on your machine; after a one-time model download, transcription runs locally on the CPU.

Get started View source

Project snapshot · measured 2026-10-26

Key facts

Commits 704Last commit 2026-10-26 Source receipt ↗ Commits · 7 days 24 Source receipt ↗ Tickets closed 97 / 178 Source receipt ↗ Studies 16 Source receipt ↗

12-week commit activity

2026-10-26

Aug 3

Aug 10

Aug 17

Aug 24

Aug 31

131

Sep 7

Sep 14

116

Sep 21

106

Sep 28

Oct 5

Oct 12

Oct 19

View 12-week data table

Week starting	Commits
2026-08-03	0
2026-08-10	0
2026-08-17	10
2026-08-24	10
2026-08-31	32
2026-09-07	131
2026-09-14	54
2026-09-21	116
2026-09-28	106
2026-10-05	66
2026-10-12	29
2026-10-19	23

Key Features

Continuous local speech recognition with physical modifier gating and Wayland keystroke injection.
GNOME Shell recording controls and a terminal resource and latency monitor.
Purpose-based audio sample storage, transcript corrections, and resilient ASR doctor fallbacks.

Verified Engineering Milestones

Purpose-based sample store (#170)
Tested models catalog and crispasr bench runner (#133, #199)
Robust ASR doctor checks and self-healing fallbacks (#197, #198, 4e453ec)

Stack

Go 1.26.5, Cobra, YAML specs, Linux/Wayland, evdev, PipeWire/ALSA, systemd, dotool, crispasr/Cohere Transcribe, optional Whisper, JavaScript/CSS GNOME Shell extension

Recorded demo

Voice input in action

Start dictation with a desktop shortcut. Voxi listens for speech, transcribes locally, and types each sentence into the active window.

How it works

Dictation that stays out of your way

Continuous eager streaming

A circular pre-roll buffer preserves the first sounds. Voxi segments speech on natural pauses and transcribes phrases in short bursts.

Hotkey-safe typing

Kernel-level monitoring detects held Ctrl, Alt, Super and Shift keys. Typed output is buffered while a modifier is down, preventing collisions with desktop shortcuts.

Local speech recognition

The default Cohere Transcribe model runs on your CPU through crispasr. Audio remains local; the model downloads once and then works offline.

Direct Wayland input

Voxi uses dotool for synthetic keystrokes, with clipboard fallback for layouts and applications that need it.

Live resource monitor

Inspect audio levels, transcription latency, hardware load and daemon health in a btop-style terminal dashboard.

Transcript corrections

Keep local replacement rules, stop words and silence artifacts to correct recognition without sending personal vocabulary anywhere.

Screenshots

Inspect every part of a dictation session

voxi eager session showing audio source, speech detection and live transcription

voxi monitor with a live loudness meter, hardware load, transcript feed and daemon status

voxi chunks list showing duration, real-time factor, RMS energy and acceptance status — Inspect recent chunks, their acoustic gate result and transcript without recording again.

Desktop controls

Start dictation from a keyboard shortcut

Bind voxi record toggle to any desktop shortcut. On GNOME, voxi shortcut setup configures the standard Super+X binding.

GNOME Keyboard Shortcuts listing Voxi Toggle Dictation on Super+X

GNOME shortcut editor showing the voxi record toggle command and Super+X binding

Quickstart

Install and start dictating

Install the latest Linux release and configure the user services with the project installer.

curl -fsSL https://ubunatic.com/voxi/install.sh | bash

Check the background agent, then toggle recording or open the live monitor.

systemctl --user is-active voxi-agent.service
voxi record toggle
voxi monitor --watch

Requirements

Linux and Wayland components

Speech recognition

crispasr with Cohere Transcribe by default; optional local Whisper models

Audio capture

PipeWire (parec) or ALSA (arecord)

Keyboard input

dotool and dotoold, with wl-clipboard as fallback

Modifier gating

Optional voxi-modifierd system service for physical key state

Developer docs

Architecture and contribution guide · full command reference

Engineering story

Offline resilience and self-healing ASR

When an optional secondary machine used for speech recognition is offline, voxi falls back to local engines, voxi doctor --fix diagnoses and repairs the setup, and the agent switches back when the primary returns. Audio still only goes to your own hardware. Read the fallback resilience study and the sample store story.

Project details

Stack

Go 1.26.5, Cobra, YAML specs, Linux/Wayland, evdev, PipeWire/ALSA, systemd, dotool, crispasr/Cohere Transcribe, optional Whisper, JavaScript/CSS GNOME Shell extension

Milestones

Purpose-based sample store (#170) · Tested models catalog and crispasr bench runner (#133, #199) · Robust ASR doctor checks and self-healing fallbacks (#197, #198, 4e453ec)

Source

https://codeberg.org/ubunatic/voxi ↗

Architecture

Layer composition

Terminal: Cobra CLI, resource monitor, Braille energy sparklines
Desktop: Linux, Wayland, GNOME Shell, evdev, PipeWire/ALSA, systemd, dotool
Inference / AI: Local CPU ASR via crispasr/Cohere Transcribe; optional Whisper via CPU/Vulkan
Containers / VM: —
Web: JavaScript/CSS website

Engineering story

voxi: the story behind the project

Voxi started with a practical desktop problem: speaking a sentence is often easier than typing it, but dictation should feel immediate and should not send private audio to a cloud service by default. Voxi is a standalone Go application for…

Read the engineering story

Measurement source · 2026-10-26

Verification receipts

Commits: 704 · Last commit 2026-10-26
git -C <repo> rev-list --count HEAD; git -C <repo> rev-list --count --since=<ISO-week-Monday> --until=<ISO-week-Sunday> HEAD; git -C <repo> rev-list --count --since=<measured-date-6-days>T00:00:00 --until=<measured-date>T23:59:59 HEAD; git -C <repo> log -1 --format=%cs HEAD; ls <repo>/issues | grep '^[0-9].*\.md$'; grep -l -E '^\*\*Status\*\*: *(Closed|Done|Resolved|Fixed|Complete)' <repo>/issues/*.md | wc -l; ls <repo>/docs/studies/*.md
repositories behind spec project ids in ~/projects; private-repo filter is a safety net and matches no spec project today; missing repos reported separately; seven calendar days ending on measured date in the measuring host's local timezone; commit counts include all authors and merge commits reachable from HEAD · measured 2026-10-26
Commits · 7 days: 24
git -C <repo> rev-list --count HEAD; git -C <repo> rev-list --count --since=<ISO-week-Monday> --until=<ISO-week-Sunday> HEAD; git -C <repo> rev-list --count --since=<measured-date-6-days>T00:00:00 --until=<measured-date>T23:59:59 HEAD; git -C <repo> log -1 --format=%cs HEAD; ls <repo>/issues | grep '^[0-9].*\.md$'; grep -l -E '^\*\*Status\*\*: *(Closed|Done|Resolved|Fixed|Complete)' <repo>/issues/*.md | wc -l; ls <repo>/docs/studies/*.md
repositories behind spec project ids in ~/projects; private-repo filter is a safety net and matches no spec project today; missing repos reported separately; seven calendar days ending on measured date in the measuring host's local timezone; commit counts include all authors and merge commits reachable from HEAD · measured 2026-10-26
Tickets closed: 97 / 178
git -C <repo> rev-list --count HEAD; git -C <repo> rev-list --count --since=<ISO-week-Monday> --until=<ISO-week-Sunday> HEAD; git -C <repo> rev-list --count --since=<measured-date-6-days>T00:00:00 --until=<measured-date>T23:59:59 HEAD; git -C <repo> log -1 --format=%cs HEAD; ls <repo>/issues | grep '^[0-9].*\.md$'; grep -l -E '^\*\*Status\*\*: *(Closed|Done|Resolved|Fixed|Complete)' <repo>/issues/*.md | wc -l; ls <repo>/docs/studies/*.md
repositories behind spec project ids in ~/projects; private-repo filter is a safety net and matches no spec project today; missing repos reported separately; seven calendar days ending on measured date in the measuring host's local timezone; commit counts include all authors and merge commits reachable from HEAD · measured 2026-10-26
Studies: 16
git -C <repo> rev-list --count HEAD; git -C <repo> rev-list --count --since=<ISO-week-Monday> --until=<ISO-week-Sunday> HEAD; git -C <repo> rev-list --count --since=<measured-date-6-days>T00:00:00 --until=<measured-date>T23:59:59 HEAD; git -C <repo> log -1 --format=%cs HEAD; ls <repo>/issues | grep '^[0-9].*\.md$'; grep -l -E '^\*\*Status\*\*: *(Closed|Done|Resolved|Fixed|Complete)' <repo>/issues/*.md | wc -l; ls <repo>/docs/studies/*.md
repositories behind spec project ids in ~/projects; private-repo filter is a safety net and matches no spec project today; missing repos reported separately; seven calendar days ending on measured date in the measuring host's local timezone; commit counts include all authors and merge commits reachable from HEAD · measured 2026-10-26
12-week commit activity
git rev-list --count --since=<Monday> --until=<Sunday> HEAD
12 latest complete ISO weeks; weekly values are from spec/metrics.yaml.

voxi types what you say, as you say it.

Key facts

12-week commit activity

Key Features

Verified Engineering Milestones

Stack

Voice input in action

Dictation that stays out of your way

Continuous eager streaming

Hotkey-safe typing

Local speech recognition

Direct Wayland input

Live resource monitor

Transcript corrections

Inspect every part of a dictation session

Eager dictation

Resource monitor

Chunk diagnostics

Start dictation from a keyboard shortcut

GNOME custom shortcut

Install and start dictating

Linux and Wayland components

Offline resilience and self-healing ASR

Project details

Layer composition

voxi: the story behind the project