Nextalk

High-performance offline voice input for Linux

Convert speech to text in real-time, input to any application via the Fcitx5 input method framework. Runs completely offline to protect your privacy.

Features

Offline Recognition - Dual-engine ASR based on Sherpa-onnx: SenseVoice offline engine (default — high accuracy, multilingual, auto punctuation) and Zipformer streaming engine (optional — real-time, character by character); data never leaves your device
Low Latency - Streaming engine transcribes as you speak; text commit path latency < 20ms
Transparent Floating Window - Borderless capsule UI with breathing animation, minimal workflow disruption
Native Wayland Support - System shortcuts and text submission both support Wayland
Bilingual UI - Simplified Chinese / English interface, switchable from the tray menu
Engine & Model Options - Hot-switch engines and Zipformer int8/standard variants from the tray menu

Quick Start

Installation

Ubuntu/Debian:

# Recommended: apt installs dependencies automatically
sudo apt install ./nextalk_0.2.13-1_amd64.deb

If you install with sudo dpkg -i, dpkg does not resolve dependencies; run sudo apt -f install afterwards if it complains about missing ones.

Fedora/CentOS/RHEL:

# Recommended: dnf installs dependencies automatically
sudo dnf install ./nextalk-0.2.13-1.x86_64.rpm

Runtime dependencies:

Dependency	Notes
`fcitx5` (≥ 5.0)	Required, text commit channel (declared by deb/rpm, installed automatically by the package manager)
`libgtk-3-0` / `gtk3`	Required (installed automatically by the package manager)
PulseAudio or PipeWire (`pipewire-pulse`)	Audio capture, shipped by default on mainstream distros
`xdg-desktop-portal` + your desktop's backend	Needed for shortcut auto-registration (bundled with GNOME 48+/KDE 5.27+, no manual install)
`libayatana-appindicator3-1` / `libayatana-appindicator-gtk3`	Tray runtime library (declared by deb/rpm since 0.2.13, installed automatically)
`gnome-shell-extension-appindicator`	GNOME only: extension required to show the tray icon; enable it and re-log after installing (see FAQ); KDE supports the tray natively

Fcitx5 will automatically restart after installation to load the plugin.

GNOME (ibus) distros (e.g. Fedora):

On GNOME desktops that default to ibus (such as Fedora), fcitx5 is not the session input method, so recognized text is committed directly through the bundled GNOME Shell extension. Enable it and log out and back in for it to take effect:

gnome-extensions enable nextalk@gonewx.com

Without the extension enabled, the app automatically falls back to clipboard mode (see Non-Fcitx5 Environment). Fcitx5 users need no action.

Note: the extension only works in Wayland sessions for native Wayland windows; when the focused app is X11/XWayland (or in an Xorg session), the app automatically falls back to clipboard mode — this is expected behavior.

Configure Hotkey

Works out of the box (supported environments): On KDE Plasma 5.27+, GNOME 48+, and Hyprland, the app auto-registers a global shortcut (default Super+Z) via XDG Desktop Portal on first launch. The system shows a one-time authorization dialog — confirm it and the shortcut takes effect immediately, no manual setup required.

Fallback (manual configuration): On environments without Portal GlobalShortcuts support (GNOME <48, wlroots/Sway, and the default sessions of Ubuntu 22.04/24.04), the app silently falls back to the system shortcut. Configure it manually and bind the nextalk-toggle command:

GNOME:

Settings → Keyboard → View and Customize Shortcuts → Custom Shortcuts
Click "Add Shortcut"
Name: Nextalk Voice Input
Command: nextalk-toggle
Shortcut: Press Super+Z (recommended; do NOT use Alt+Space — GNOME's window-menu key already occupies it)

KDE Plasma:

System Settings → Shortcuts → Custom Shortcuts
Edit → New → Global Shortcut → Command/URL
Trigger: Set to Super+Z
Action: nextalk-toggle

The current hotkey mode (Portal / system) is shown as a read-only item in the tray menu.

Usage

Launch App - Start Nextalk from app menu, or run nextalk
Press Hotkey - Floating window appears and starts recording
Speak - See recognized text in real-time
Press Hotkey Again - Stop recording, text auto-inputs to current app
Or Wait for Auto-submit - Text submits automatically after pause

First Run: The app will automatically download the speech recognition model (~200MB)

Command Line Options

Option	Description
`--toggle`	Toggle recording state (for system hotkey)
`audio`	Manage audio input device (interactive mode)
`audio <number>`	Set audio device by index number directly
`audio --list`	List available devices (machine-readable)
`audio default`	Reset to system default device
`--help`	Show help information
`--version`	Show version

Non-Fcitx5 Environment

When Fcitx5 is not running, GNOME desktops can commit text directly via the bundled Shell extension (see GNOME (ibus) distros above); if the extension is unavailable (or the focused app is X11/XWayland), the app automatically uses clipboard mode:

Recognized text is copied to system clipboard
UI shows "Copied to clipboard" prompt
Manually paste (Ctrl+V) to target application

System Tray

The app supports system tray (on supported desktop environments), right-click menu provides:

Show/Hide window
Reconnect Fcitx5 - manually reconnect when the input-method channel drops
Hotkey mode (read-only) - shows the active hotkey path: Hotkey: Super+Z (Portal auto-registered) or Hotkey: System settings (nextalk-toggle)
Model settings - switch ASR engine (SenseVoice / Zipformer) and Zipformer model version (int8 / standard)
Switch UI language (中文 / English)
Audio input device - Select from available audio input devices
Settings - open the config directory
Exit app

Note: On GNOME the tray icon only shows after installing and enabling the gnome-shell-extension-appindicator extension (KDE supports it natively) — see the FAQ. Without a tray the app still works via the nextalk --toggle command.

System Requirements

Component	Requirement
OS	Linux (Ubuntu 22.04+ recommended)
Display Server	X11 or Wayland
Audio System	PulseAudio or PipeWire (`pipewire-pulse`)
Input Method	Fcitx5

Configuration

Engine & Model Settings

Switch via system tray menu:

Engine	Description
`sensevoice` (default)	Offline engine: recognizes per VAD segment, higher accuracy, auto punctuation, multilingual (zh/en/ja/ko/yue)
`zipformer`	Streaming engine: recognize-while-listening, text appears character by character

Zipformer variants: int8 (faster, smaller memory) / standard (higher accuracy).

Config File

Advanced configuration: ~/.config/nextalk/settings.yaml

model:
  # ASR engine: zipformer | sensevoice
  engine: sensevoice

  zipformer:
    type: int8        # Model variant: int8 | standard
    custom_url: ""    # Custom model download URL (empty = default)

  sensevoice:
    use_itn: true     # Inverse text normalization
    language: auto    # auto | zh | en | ja | ko | yue
    custom_url: ""

audio:
  input_device: "default"  # Audio input device: "default" or device name

Audio Device Selection:

Use "default" to use system default audio device
Specify device name (e.g. "Built-in Audio Analog Stereo") to use a specific device
Use nextalk audio command or system tray menu to select devices interactively

Hotkey Configuration: On supported environments the hotkey is auto-registered via XDG Desktop Portal (no config needed); otherwise it is configured through your desktop environment's native settings, not through this config file. See Configure Hotkey section for details.

Build from Source

Prerequisites

# Ubuntu/Debian
sudo apt install fcitx5 fcitx5-dev libportaudio2 portaudio19-dev cmake build-essential

# Install Flutter: https://flutter.dev/docs/get-started/install/linux

Local Build

# Using Makefile (recommended)
make build

# Or build separately
make build-flutter    # Flutter client
make build-addon      # Fcitx5 plugin

Docker Build (Cross-distro Compatible)

Recommended to use Docker build for cross-distribution compatibility:

# Incremental build (recommended)
make docker-build

# Full rebuild
make docker-rebuild

# Flutter only
make docker-build-flutter

# Plugin only
make docker-build-addon

Install Plugin

make install-addon          # User-level installation
make install-addon-system   # System-level installation (requires sudo)

Run

make run          # Development mode
make run-release  # Release version

Build Packages

./scripts/build-pkg.sh --deb   # DEB package
./scripts/build-pkg.sh --rpm   # RPM package
./scripts/build-pkg.sh --all   # All formats

Output in dist/ directory.

📦 For the full maintainer release flow, CI/CD automation, and version management, see the Build and Release Guide.

Uninstall

Ubuntu/Debian:

sudo dpkg -r nextalk

Fedora/CentOS/RHEL:

sudo rpm -e nextalk

User data is preserved at ~/.local/share/nextalk/, delete manually for complete cleanup.

Project Structure

nextalk/
├── voice_capsule/        # Flutter client
│   └── lib/
│       ├── main.dart     # Entry point
│       ├── ffi/          # FFI bindings (sherpa, portaudio)
│       ├── services/     # Business logic
│       │   ├── asr/      # ASR engine abstraction
│       │   └── ...       # Other services
│       ├── ui/           # Widget components
│       └── l10n/         # Internationalization
├── addons/fcitx5/        # Fcitx5 C++ plugin
├── docs/                 # Design documents
├── scripts/              # Build scripts
├── libs/                 # Precompiled dynamic libraries
├── packaging/            # DEB/RPM templates
└── Makefile              # Build entry

Detailed architecture: docs/architecture.md

Development

make test       # Run tests
make analyze    # Code analysis
make clean      # Clean build
make help       # Show all commands

FAQ

System Tray Icon Not Showing (GNOME/Fedora)

GNOME desktop doesn't show AppIndicator tray icons by default; install and enable the gnome-shell-extension-appindicator extension:

# Debian/Ubuntu
sudo apt install gnome-shell-extension-appindicator

# Fedora
sudo dnf install gnome-shell-extension-appindicator

Log out and back in after installing (a Wayland session must be re-logged to load the extension), then make sure it is enabled:

gnome-extensions enable appindicatorsupport@rgcjonas.gmail.com

Restart Nextalk and the tray icon appears. KDE Plasma supports the tray natively with no configuration.

Without the extension the tray is invisible, but the app still works fully via the command line:

nextalk --toggle  # Toggle recording state

App Crashes on Startup (Segfault)

0.2.13 fixes the startup segfault on Fedora caused by the legacy libappindicator library (the tray plugin now prefers the ayatana implementation). If you still hit a tray-related crash, temporarily disable the tray to isolate it:

NEXTALK_NO_TRAY=1 nextalk

When configuring system shortcut (fallback mode):

nextalk-toggle

Hotkey Not Responding

Confirm hotkey is configured in system settings (command: nextalk-toggle)
Confirm Nextalk app is running (check system tray)
Test command line: nextalk --toggle

Text Not Inputting to App

Ensure Fcitx5 plugin is correctly installed:

ls ~/.local/lib/fcitx5/libnextalk.so
ls ~/.local/share/fcitx5/addon/nextalk.conf
fcitx5 -r  # Restart Fcitx5

If not using Fcitx5, the app will automatically use clipboard mode.

Model Download Failed

Configure custom download URL or use proxy:

# ~/.config/nextalk/settings.yaml
model:
  zipformer:
    custom_url: "https://your-mirror/zipformer-model.tar.bz2"
  sensevoice:
    custom_url: "https://your-mirror/sensevoice-model.tar.bz2"

Audio Device Issues

If you encounter "Audio device busy" error (e.g., when using EasyEffects), you can select a different audio input device:

Method 1: CLI Interactive Mode

nextalk audio       # Enter interactive mode to select device

Method 2: CLI Direct Selection

nextalk audio --list  # List available devices
nextalk audio 2       # Select device by index
nextalk audio default # Reset to system default

Method 3: System Tray Right-click tray icon → Audio Input Device → Select from available devices

Debug Commands:

nextalk audio --list  # List devices detected by the app
pactl list sources    # System-level debug command

Contributing

Contributions, bug reports, and suggestions are welcome!

Fork this repository
Create feature branch (git checkout -b feature/amazing-feature)
Commit changes (git commit -m 'Add amazing feature')
Push branch (git push origin feature/amazing-feature)
Create Pull Request

Acknowledgments

Sherpa-onnx - High-performance offline speech recognition engine
Fcitx5 - Modern input method framework
Flutter - Cross-platform UI framework

License

MIT License

Issue Feedback: GitHub Issues

Name		Name	Last commit message	Last commit date
Latest commit History 115 Commits
.agent/workflows/bmad		.agent/workflows/bmad
.claude		.claude
.codegraph		.codegraph
.github/workflows		.github/workflows
_bmad-output		_bmad-output
addons		addons
docker		docker
docs		docs
libs		libs
packaging		packaging
scripts		scripts
third_party/system_tray		third_party/system_tray
voice_capsule		voice_capsule
.gitignore		.gitignore
.mcp.json		.mcp.json
CLAUDE.md		CLAUDE.md
Makefile		Makefile
README.md		README.md
README_zh.md		README_zh.md
version.yaml		version.yaml

Folders and files

Latest commit

History

Repository files navigation