Karpender

Karpender is a Fedora/Linux desktop app for voice privacy. It reads a physical microphone through PipeWire, applies voice anonymization DSP, and exposes the result as a virtual microphone for apps such as Telegram, Discord, OBS, browsers, and Zoom.

physical mic -> Karpender DSP -> Karpender Privacy Voice Mic -> chat/recording app

Features

• GTK4/libadwaita desktop UI.
• PipeWire input source selection.
• Virtual microphone output.
• Voice anonymization with pitch/timbre/prosody changes.
• Noise Cleanup as a soft expander, not a hard gate.
• Privacy Amount for stronger or lighter voice disguise.
• Flatten Intonation mode for reducing recognizable prosody.
• Monitor to Speakers mode for hearing the processed voice locally.

Use headphones when testing monitor mode to avoid feedback.

Requirements

Runtime:

• Fedora or another PipeWire-based Linux desktop.
• PipeWire and WirePlumber running.
• GTK4 and libadwaita runtime libraries.
• pw-dump available in PATH for input device discovery.

Development packages on Fedora:

sudo dnf install rust cargo gtk4-devel libadwaita-devel pipewire-devel clang pkgconf-pkg-config jq

Run From Source

cargo run

In Karpender:

1. Select the real microphone in Input Device.
2. Adjust Privacy Amount, Noise Cleanup, and Flatten Intonation.
3. Press Start Processing.
4. In Telegram/Discord/browser, select Karpender Privacy Voice Mic as the microphone.

AppImage Build

The AppImage setup lives in:

• scripts/build-appimage.sh
• packaging/appimage/AppRun
• data/com.github.xierongchuan.karpender.desktop
• data/com.github.xierongchuan.karpender.metainfo.xml
• data/icons/com.github.xierongchuan.karpender.svg

Recommended tooling in PATH:

• linuxdeploy
• linuxdeploy-plugin-gtk

Fallback:

• appimagetool

Build:

scripts/build-appimage.sh

This command only builds local artifacts in target/appimage and dist. It does not install Karpender into the system. If neither linuxdeploy nor appimagetool is found in PATH, the script downloads a local portable appimagetool into target/appimage-tools and uses it from there.

Output:

dist/Karpender-<version>-<arch>.AppImage

If no AppImage builder can be found or downloaded, the script still prepares the AppDir at:

target/appimage/Karpender.AppDir

Then put appimagetool in PATH or at target/appimage-tools/appimagetool-<arch>.AppImage and rerun the script.

Flatpak Build

The Flatpak setup lives in:

• scripts/build-flatpak.sh
• scripts/generate-flatpak-cargo-sources.py
• packaging/flatpak/com.github.xierongchuan.karpender.json
• packaging/flatpak/cargo-sources.json

Required tooling:

• flatpak
• flatpak-builder
• Flathub configured for the user Flatpak installation. scripts/build-flatpak.sh adds the user remote automatically if it is missing.

Build a local bundle:

scripts/build-flatpak.sh

The script regenerates packaging/flatpak/cargo-sources.json from Cargo.lock, builds against org.gnome.Platform//50 and the Rust/LLVM SDK extensions, writes a local OSTree repository under target/flatpak, and exports:

dist/Karpender-<version>.flatpak

Install the exported bundle locally:

flatpak install --user dist/Karpender-<version>.flatpak
flatpak run com.github.xierongchuan.karpender

The manifest grants Wayland/X11 fallback, GPU, PulseAudio, and PipeWire socket access so Karpender can show the GTK UI, read microphone nodes, create the virtual microphone, and monitor processed audio.

Development Checks

cargo fmt --check
cargo check
cargo test

PipeWire Debugging

List PipeWire nodes:

pw-dump --no-colors

Compact node view:

pw-dump --no-colors | jq -r '.[] | select(.type=="PipeWire:Interface:Node") | [.id, .info.props["media.class"], .info.props["node.name"], .info.props["node.description"], .info.state] | @tsv'

Expected flow while running:

• real microphone node is selected by Karpender;
• Karpender Privacy Voice Mic appears as an Audio/Source;
• Telegram/Discord/browser links to the virtual source.

Notes

Karpender aims to make voice recognition harder while keeping speech understandable. It is not a guarantee of anonymity against forensic voice analysis.