Repository navigation
SoundVisualizer_v2.0
SoundVisualizer 2.0 brings over what the Android version learned: a new AI pipeline, a Sounds tab, a lighter overlay, help and notices, a tutorial, better keyboard and screen reader support, and 18 languages.
Download
SoundVisualizer_v2.0.zip — Windows 10/11 (64-bit). You no longer need to install .NET. Extract the zip anywhere and run SoundVisualizer.exe. Settings are saved in settings.json next to the exe; to keep your v1.x settings, copy your old settings.json into the new folder.
What's new
AI sound classification, rebuilt
- The audio now reaches YAMNet the way the model was trained: anti-aliased resampling to 16 kHz and the exact log-mel frontend the model was exported with. Each step is checked against reference outputs.
- On 5.1/7.1 systems the AI listens to every channel, not just the front.
- Warning sounds (sirens, alarms, explosions, fireworks and similar) count as Danger even when they are only among the top five candidates, so music or game audio can't hide them. Gunshot labels are not promoted this way, to avoid false alarms.
- Colors no longer flicker: Danger shows on the first qualifying result, Speech and Ambient need two in a row.
- The AI runs every 250 ms and rests while it's quiet.
- The separate gunshot booster model was removed.
Sounds tab
- Choose whether each group of sounds (or, in advanced mode, each of the 521 sounds) counts as Ambient, Speech or Danger. Search, filter, and reset in one step. The overlay follows your changes immediately.
A smoother, lighter overlay
- Animations move at the same speed at any frame rate, from 30 FPS up to your monitor's refresh rate.
- When nothing has been visible for a second, the overlay stops redrawing every frame and wakes as soon as sound arrives (idle CPU use dropped to roughly a third in our measurement).
- A Danger result that arrives a moment late is still drawn at the size the sound had.
- With stereo output, sound common to both channels is drawn at the top and only the left/right difference goes to the sides.
- Fixed: Wave and Outline drew nothing in 2.0 mode on 16:9 screens, and disappeared in 5.1/7.1 whenever a corner channel was silent.
Help, notices, tray icon, and tutorial
- A new Help tab with getting started, visual modes, sound types, hotkeys, FAQ, a report button (prefilled GitHub issue or email), and the open source licenses.
- The Home tab tells you when there is no playback device to capture, when sound type detection is unavailable, or why the visualizer stopped.
- A tray icon to start or stop the visualizer, return to the launcher, or exit.
- A short tutorial on first launch, which you can reopen from the Home tab.
Look and language
- 18 languages: Korean, English, Japanese, Chinese (Simplified and Traditional), Vietnamese, Thai, Indonesian, Spanish, Portuguese (Brazil), French, Italian, German, Russian, Polish, Turkish, Hindi, and Arabic (right-to-left). By default the app follows the Windows display language.
- A refreshed design matching the Android version, with the Pretendard font.
Keyboard and screen readers
- Every tab and window can now be reached with the keyboard (Tab used to reach only the last tab's content). Ctrl+Tab switches launcher tabs and Esc closes dialogs and the overlay settings.
- Browsing the language list with the keyboard no longer changes the language until you choose one.
- Buttons and controls that were silent for screen readers now have names (minimize, close, hotkeys, color buttons, the color picker area and presets), status changes and notices are announced, and section titles are exposed as headings.
- Everything you can drag in the overlay editor can also be changed with a slider in its panel.
Developer mode
- Shows what the AI heard (top five sounds, confidence, timing) at the top left of the overlay, and can record every result to a CSV file in the
ai-logfolder next to the app.
Other fixes
- The overlay now uses the language you chose (six languages used to start in Korean).
- VB-CABLE routing respects the actual buffer length, and 7.1 channels are mapped in the right order.
- "Reset to defaults" now also resets the hotkeys.
- Settings are saved once after you stop moving a slider instead of on every step.
- Switching the visual mode in the overlay settings now shows that mode's own size, opacity, and speed values.
- The app name is now written as one word, SoundVisualizer, everywhere.
Notes
- Sound type detection can be wrong. Don't rely on this app as your only safeguard when safety is at stake.
- Licenses for bundled libraries, the YAMNet model, the AudioSet sound names (CC BY-SA 4.0), the Pretendard font (OFL), and the tutorial sounds (CC0 / public domain) are in
THIRD_PARTY.mdand in the app under Help → About → Open source licenses.