Omahub
← All plugins
H

Voxtype Enhance

by HANCORE

A native Omarchy control center for Voxtype dictation with model downloads, language, and terminal-aware universal paste controls.

GitHub

Security review

No obvious issues detected

Deterministic scan — not a security guarantee

None
Risk level
None
Analyzed commit
47226f6
Scanned
1 week ago

No potentially dangerous behavior detected in the analyzed commit.

Automated analysis only — not a security guarantee.

AI advisory review

Review recommended

Language-model assessment · ~deepseek/deepseek-v4-flash-latest — advisory only

Low
AI risk level
Low
Recommendation
review
Model
~deepseek/deepseek-v4-flash-latest
Analyzed commit
47226f6
Reviewed
1 week ago

The plugin is a transparent Quickshell widget backed by Python scripts with no obfuscation or hidden persistence. It can perform a few root-level operations (ONNX enablement, optional voxtype-bin install, ARM ONNX binary install) and modifies Voxtype config/user service, but these are user-initiated, fixed commands, and documented. The main independent concern is that the bar widget runs `hyprctl eval 'hl.unbind("mouse:274")'` at startup even when the middle-mouse toggle is disabled, which can unexpectedly remove an existing Hyprland binding.

  • bar/widget.qml applies `hl.unbind("mouse:274")` on startup even with middleMouseToggle false, contradicting the README's 'off by default to avoid conflicts' and potentially removing a user's existing middle-button binding.
  • The Python bridge can elevate via pkexec: `voxtype setup onnx --enable`, `pacman -S --noconfirm --needed voxtype-bin`, and an ARM helper that installs a verified binary to `/usr/local/bin/voxtype`; these are explicit and disclosed but are real system-level changes.
  • ARM ONNX support relies on a pinned GitHub release URL and an embedded SHA-256; the root install helper is only as trustworthy as that pin and the approval flow.
  • The widget/panel invokes `hyprctl` and can rewrite `~/.config/voxtype/config.toml` and restart the user voxtype service; expected behavior, but worth confirming it does not conflict with user configs.
How this check works

This review combines the deterministic scan (the rule-based results above) with an independent look at the plugin's code by a language model. The model reads a trimmed sample of the repository's files, the manifest, and the README, then gives a plain-language risk level and a recommendation: install (no notable danger), review (look closer first), or avoid (clearly dangerous).

It runs on the same analyzed commit as the deterministic scan and is strictly advisory — it is not a security guarantee and never blocks a plugin by itself. A human moderator still reviews plugins before they are listed.

AI advisory only — automated analysis, not a security guarantee.

Install
$ omarchy plugin add https://github.com/iamcheyan/omarchy-voxtype-enhance --enable
Productivity #bar #quickshell #ai

Voxtype Enhance

0.1.1

  • Adapted model, language, output, status, and border colors to Omarchy popup theme tokens for readable contrast across themes.

Voxtype Enhance is an Omarchy experience-enhancement plugin for Voxtype voice input. It adds a native top-bar microphone control and a compact settings panel for speech models, language, and output behavior.

Screenshot

Voxtype Enhance settings panel

English: The settings panel is opened from the microphone icon in the top bar. It shows the available speech models, language choices, and output modes.

中文： 点击顶栏麦克风图标即可打开输入法设置面板。面板提供语音模型、识别语言和输出方式设置。

日本語： トップバーのマイクアイコンから音声入力設定パネルを開けます。音声モデル、認識言語、出力方式を設定できます。

Features

  • Native Omarchy top-bar microphone indicator.
  • The microphone icon gently breathes from its theme color toward yellow while Voxtype is recording, then returns to its normal appearance during transcription and when idle.
  • Recording animation can be enabled or disabled from the Voxtype settings panel.
  • Three Sasayaki-compatible offline speech models, with engine selection handled automatically.
  • Automatic model download, installation, file-size checking, and SHA-256 verification.
  • Chinese, English, Japanese, Korean, and automatic language choices.
  • Omarchy universal paste: Shift+Insert in terminal/TUI windows and Ctrl+V in graphical applications.
  • Voxtype's native Type output as an alternative.
  • Real-time download progress shown directly inside the selected model card.
  • Language and output controls stay hidden until at least one model is downloaded.
  • Clear plugin data action for resetting model downloads and plugin-managed settings.
  • No replacement recorder, speech recognizer, clipboard daemon, or extra Quickshell service.

Installation

Install Voxtype first through Omarchy's AI / Dictation setup, then install the plugin:

omarchy plugin add https://github.com/iamcheyan/omarchy-voxtype-enhance.git --enable

The plugin provides a bar-widget entry point. If the microphone icon is not automatically placed in the top bar, add Voxtype Enhance to the right side of the bar layout.

The settings panel includes an optional middle-mouse dictation toggle. It is off by default to avoid conflicts with existing user bindings. When enabled, the first click on middle mouse (mouse:274) starts Voxtype recording and the second click stops it and transcribes. The binding consumes the middle-button event, so applications do not receive it as the usual primary-selection paste. F9 and the existing keyboard shortcuts remain available.

Models

The panel shows model names rather than engine names. Selecting a model downloads missing files, verifies them, installs them into Voxtype's model directory, and then switches the engine:

Panel option Internal engine Voxtype model Size
SenseVoice Small · int8 sensevoice small-int8 229 MB
SenseVoice Small · full precision sensevoice small 894 MB
Paraformer Large · int8 paraformer paraformer-zh 232 MB

Only the model confirmed as active in Voxtype is highlighted. A downloaded model that is not active is labelled downloaded, not active; a failed engine switch shows the Voxtype/systemd error and restores the effective selection. Downloads use the pinned Sasayaki Hugging Face sources and SHA-256 manifests. Already verified files are skipped. Progress is streamed line-by-line while files are being downloaded, so the model card reflects the current transfer instead of remaining at the initial 0% state.

SenseVoice and Paraformer require an ONNX-capable Voxtype binary. The default Voxtype installation uses the standard Whisper binary. When an ONNX model is selected without ONNX support, the plugin stops before downloading and explains what is missing and provides an Enable ONNX support (administrator approval required) action in the panel. On x86_64, the user can explicitly approve the fixed command pkexec voxtype setup onnx --enable. On aarch64/ARM, the plugin instead offers to download and SHA-256 verify Voxtype's pinned official ARM ONNX release, then installs it under /usr/local/bin and uses a per-user systemd override; it does not replace the package-owned /usr/bin/voxtype. Only after the user approves the privileged install does the model download resume. Cancelling leaves the existing setup unchanged.

If Voxtype itself is missing (for example when only the plugin was installed), any panel action that changes settings first offers to install Omarchy's voxtype-bin package through the fixed command pkexec pacman -S --noconfirm --needed voxtype-bin. Cancelling the authentication prompt leaves the system unchanged; read-only status queries never trigger an installation.

On NixOS, the plugin never invokes pacman. Install Voxtype declaratively in the NixOS configuration first, for example with voxtype-onnx, then rebuild the system. The panel reports this route instead of attempting an imperative package install.

Models are stored under:

~/.local/share/voxtype/models/

Output modes

The plugin intentionally exposes two output modes:

Omarchy universal paste

Voxtype writes the complete transcription to the Wayland clipboard. The plugin then detects the focused Hyprland window using the same terminal-tag policy as Omarchy's clipboard.lua:

  • Omarchy terminal/TUI windows receive Shift+Insert;
  • graphical applications receive Ctrl+V.

This avoids terminal applications interpreting Ctrl+V as a special action, such as an image-paste command. The final shortcut is sent through Hyprland's hl.dsp.send_key_state dispatcher with explicit key press and release events.

Before output, the plugin records the previous clipboard hash. It sends the shortcut only when Voxtype has produced new non-empty text, so pressing the recording key without speaking does not paste stale clipboard content.

Type

Voxtype's native simulated keyboard-input mode. This can be useful in applications where direct typing is preferred, but mixed-language text may interact with Fcitx5/Rime character by character.

Settings and reset

The panel updates ~/.config/voxtype/config.toml only after the user selects a setting. Changes restart the existing user-level voxtype.service so Voxtype loads the new engine or output configuration.

The Clear plugin data link removes only the three model directories managed by this plugin and restores SenseVoice int8, Chinese, and Omarchy universal paste. It does not remove Whisper models, recordings, hotkeys, or unrelated Voxtype data. After clearing, choose a model to download it again.

The plugin does not install a system service. When Voxtype is already installed, model selection does not require privileges. Exactly three operations elevate through pkexec; all are visible, user-initiated, and run fixed commands:

  • pkexec voxtype setup onnx --enable — switch to the ONNX-capable binary;
  • the pinned official ARM ONNX binary — offered only after explicit approval on aarch64/ARM, installed outside the package manager and verified by SHA-256;
  • pkexec pacman -S --noconfirm --needed voxtype-bin — offered only when the Voxtype binary is missing entirely.

When no model is downloaded, the panel hides the Language and Output controls until the first model has been verified successfully. This prevents settings from being applied while Voxtype has no usable transcription model.

Dependencies

The following are provided by Omarchy, Voxtype, or the normal desktop setup:

  • Omarchy Quickshell;
  • Voxtype and its user service;
  • Python 3 standard library;
  • Hyprland and hyprctl for universal paste;
  • wl-copy for clipboard output;
  • Voxtype's normal output dependencies, such as wtype.

Model downloads require network access only when a selected model is not already installed.

Uninstallation

Before disabling or removing the plugin, switch Output to a non-universal mode once so the plugin removes its pre_output_command and post_output_command hooks from ~/.config/voxtype/config.toml:

python3 scripts/voxtype-config.py set mode type

Then disable or remove the plugin using Omarchy's plugin manager. Removing the plugin removes the top-bar control and plugin files. It does not remove Voxtype, user recordings, or downloaded models; use Clear plugin data before removal if those model files should also be deleted.

Validation

From the plugin repository root:

omarchy plugin validate .
/usr/lib/qt6/bin/qmllint -I /usr/share/omarchy/shell \
  bar/widget.qml VoxtypePanel.qml
python3 -m py_compile scripts/*.py
python3 -m unittest discover -s tests -v

License

MIT. See LICENSE.

中文说明

Voxtype Enhance 是一个 Omarchy 体验增强插件，用于增强 Voxtype 语音输入。它在顶栏提供麦克风图标和设置面板，让用户可以管理语音模型、识别语言和输出方式。

主要功能：

  • 顶栏麦克风状态图标；
  • 三个 Sasayaki 兼容的离线语音模型；
  • 选择模型后自动下载、校验、安装并切换 engine；
  • 中文、英文、日文、韩文和自动识别；
  • 终端使用 Shift+Insert、普通应用使用 Ctrl+V 的万能粘贴；
  • Voxtype 原生 Type 输入模式；
  • 实时模型下载进度显示；
  • 未下载模型时隐藏语言和输出设置；
  • 清除模型并恢复插件默认设置。

安装 Voxtype 后执行：

omarchy plugin add https://github.com/iamcheyan/omarchy-voxtype-enhance.git --enable

点击顶栏麦克风图标打开设置面板。用户不需要了解模型对应的 engine，选择模型后插件会自动处理。插件不会修改录音快捷键，仍然使用 Voxtype 原来的快捷键。

如果缺少 ONNX 支持，面板会显示 Enable ONNX support (administrator approval required) 操作。点击后会请求管理员权限，ARM 平台会下载并校验对应的 ONNX 版本，完成后自动继续模型下载。模型下载过程中会实时显示进度；在模型下载完成前，语言和输出设置会暂时隐藏。

万能粘贴会根据当前窗口判断输出方式：Omarchy 标记为终端的窗口使用 Shift+Insert，其它图形应用使用 Ctrl+V，以避免 CodeX 等终端程序把 Ctrl+V 解释成图片粘贴操作。

点击面板底部的 Clear plugin data 会删除插件管理的三个模型，并恢复 SenseVoice int8、中文和万能粘贴。不会删除 Whisper 模型、录音、快捷键或其它 Voxtype 数据。

日本語

Voxtype Enhance は、Voxtype の音声入力体験を Omarchy 向けに強化するプラグインです。トップバーのマイクアイコンから、音声モデル、認識言語、出力方式を設定できます。

主な機能：

  • トップバーのマイク状態表示；
  • Sasayaki 互換の 3 つのオフライン音声モデル；
  • モデル選択時の自動ダウンロード、検証、インストール、engine 切り替え；
  • 中国語、英語、日本語、韓国語、自動認識；
  • ターミナルでは Shift+Insert、GUI アプリでは Ctrl+V を使う Omarchy universal paste；
  • Voxtype 標準の Type 入力；
  • リアルタイムのモデルダウンロード進捗表示；
  • モデル未ダウンロード時は言語と出力設定を非表示；
  • ダウンロード済みモデルと設定を初期化するリセット機能。

Voxtype を先に Omarchy からインストールし、次のコマンドでプラグインを追加します。

omarchy plugin add https://github.com/iamcheyan/omarchy-voxtype-enhance.git --enable

トップバーのマイクアイコンをクリックして設定パネルを開きます。モデルの engine を意識する必要はありません。モデルを選ぶと、必要なファイルのダウンロードと設定変更が自動的に行われます。録音ショートカットは変更せず、Voxtype の既存設定を使用します。

Universal paste はフォーカス中のウィンドウを判定し、Omarchy のターミナルでは Shift+Insert、通常の GUI アプリでは Ctrl+V を送信します。これにより、CodeX などの TUI が Ctrl+V を画像貼り付けとして扱う問題を避けられます。

ONNX が必要な場合は、パネルの Enable ONNX support (administrator approval required) をクリックしてください。管理者権限を確認した後、ARM では対応する ONNX 版 Voxtype をダウンロード・検証し、モデルのダウンロードを自動的に再開します。モデルが未ダウンロードの間は、言語と出力設定を表示しません。

パネル下部の Clear plugin data をクリックすると、プラグインが管理する 3 つのモデルを削除し、SenseVoice int8、中国語、Universal paste の初期設定に戻します。Whisper モデル、録音データ、ショートカット、その他の Voxtype 設定は削除しません。