SOVIS brings the sounds you can't hear back onto the screen
Our story
We draw the sounds that captions leave out
Which way footsteps are coming from. Whether a siren is going off. Whether someone is talking to you. Hearing accessibility has grown around conveying dialogue, through captions and sign language. Sounds other than dialogue are still left out.
SOVIS is a team of students at Dongguk University. In spring 2026, we built SoundVisualizer for Windows in an open source software project course. After testing it twice with deaf and hard of hearing students, we brought the same design to phones and made the Android app.
About 440,000
People registered with a hearing disability in Korea at the end of 2024. That is 16.8% of the 2.63 million registered people with disabilities, the second-largest group after physical disabilities.
Ministry of Health and Welfare, Status of Registered Persons with Disabilities 2024
31.7%
The share of people with a hearing disability among everyone newly registered with a disability in 2024. It is the largest group among new registrations.
Ministry of Health and Welfare, Status of Registered Persons with Disabilities 2024
Captions and sign languageConvey dialogueInformation such as where a sound comes from, footsteps or sirens is lost.
Sound indicators in some gamesWork only inside that gameVideos, music and most apps and players have nothing like this.
Dedicated devicesHave to be bought separatelyGetting the equipment costs money.
SoundVisualizerWorks with any mediaDraws the sound playing on your phone or computer, from videos, music, games and more, right on the screen, whatever the app.
Who is it for?
For every moment when sound is hard to hear, or can't be turned on
01
Media access for deaf and hard of hearing people
It turns the direction and type of sounds that aren't dialogue, such as footsteps, gunshots and sirens, into something you can see, so movies, dramas, videos, music and games can be enjoyed more fully.
02
Any app, as it is
It isn't tied to one app or game. Any sound that plays on your device, from YouTube and streaming services to music and games, is drawn right over what you are watching.
03
When you can't turn the sound on
On the subway, in a lecture hall, late at night, in a room where a child is sleeping. Even when you can't play sound freely, you can see where a sound comes from and what it is.
04
Everyday sounds around you
Turn on External sound mode on Android and the app listens through the microphone. It alerts you to fire alarms, sirens or someone calling you, on screen and with vibration.
Our values
Built with the people it's for
We chose features from what deaf and hard of hearing students told us, not from our own guesses. We tested twice in May 2026 and ran a field trial of the Android app in October.
On your device
Measuring loudness and telling sound types apart all happens on your phone or computer. Sound is never saved to a file or sent to a server.
Easy for everyone
It supports 18 languages, and a tutorial shows you how to use it the first time you open it. The Windows version also works with a keyboard and screen readers.
Our journey
It started in a one-semester course and became two apps, for phones and computers
2026.03
Started building the Windows app in Dongguk University's open source software project course
2026.05
First user test at Seoul National School for the Deaf (May 19–20)
Second test after changes based on the feedback (May 29)
2026.06
Windows v1.0 released, source code published under AGPL-3.0
2026.08
Windows v1.1 released (Aug 26)
Selected as a finalist in the Digital Inclusion track of the 2026 Disability Hackathon
2026.09
Android v1.0.0 released (Sep 15)
Mid-stage build submitted to the hackathon (Sep 18)
2026.10
Android v1.6.0: External sound mode (Oct 6)
Android v1.7.2 released (Oct 13)
Windows v2.0 released (Oct 15)
Field trial of the Android app with deaf and hard of hearing students (Oct 21)
Android v1.8.0 released (Oct 22)
2026.11
Honorable Mention at the 2026 Disability Hackathon (Nov 10)
Android v1.8.4 and Windows v2.1.1 released (Nov 11)
The people behind SOVIS
Three Dongguk University students share development, AI and design
DK
Dohwan Kim
Team lead · Development · Design
Windows version
Built the on-screen overlay; the Wave, Circle and Outline modes; direction display for 5.1 and 7.1 channels with automatic output device detection; the launcher, settings, hotkeys and live edit screen; and fixed size, which shows loudness by opacity. Also tuned performance. In v2.0, added the Sounds tab, the tutorial and screen reader support, and brought over the AI classification refined on Android. In v2.1, added the Simple settings view and early display of human sounds.
Android version
Owned the app architecture and the sound capture service, overlay drawing and the four visual modes, vibration by sound type, External sound mode, the lock screen alert for Danger sounds, the Sounds tab, tutorial and help, 18 languages, testing and releases.
Also
Running the team, design, READMEs and contest documents
GP
Garam Park
AI
Windows version
Matched the YAMNet input and preprocessing to the model's specification, and built the rules that sort sounds into Ambient, Human sounds and Danger, along with smoothing so results don't flicker. Trained and tuned a helper model that catches gunshots buried in music and sound effects, and refined the 7.1-channel AI input and the Wave drawing at the corners.
Android version
Built the AI classification pipeline (log-mel preprocessing, YAMNet ONNX inference, three-type classification, the gunshot helper model) together with tests that compare it against reference output. Switched to Qualcomm's YAMNet preprocessing; made the AI's decisions take into account danger cues, human vocal sound classification and the sound types users change; and set up the evaluation tools and the 0.25-second inference interval.
Also
Guide for collecting sound data to evaluate the AI
SK
Seongjin Kim
Development
Windows version
Stabilized the capture engine that receives system audio (WASAPI loopback), and built FFT processing, routing and downmixing from 8 channels to 2, latency measurement, and the capture info display in Developer mode.
Android version
Set up the issue and PR templates, the contribution guide and branch rules, and made the app wait for any running AI inference before closing the model when the visualizer stops.
Also
Arranged the user tests with Seoul National School for the Deaf
Frequently asked questions
Questions about each device are on the Android and Windows pages
Does it record sound or send it anywhere?
No. Sound is only used on your device to measure loudness and tell sound types apart. It is never saved to a file or sent anywhere. It works without an internet connection.
Which devices does it run on?
Phones with Android 10 or later, and computers with Windows 10 or 11 (64-bit). For features and installation, see the Android and Windows product pages.
What if the AI gets a sound wrong?
Sound type detection is sometimes wrong. In the app's Sounds tab you can change which type each sound counts as, and telling us with the report button in the Help tab helps us a lot in fixing it. When safety is at stake, don't rely on this app as your only safeguard.
Does it replace captions or sign language?
No. It doesn't turn speech into text. It adds what captions and sign language can't convey: the direction, loudness and type of sounds.
Can we test it or work with you?
Yes. We're looking for institutions and organizations to run usability tests with deaf and hard of hearing people, or to advise us. Please contact us.
Contact
Build it with us
SoundVisualizer grows with feedback from the people who use it. Your impressions, sounds it drew wrong, ideas for working together: all are welcome.
We're looking for partners to run usability tests with deaf and hard of hearing people, or to advise us. We provide the install files and a user guide, and share the results with you.
Users
Tell us about sounds it got wrong or anything that gets in your way. You can report straight from the app's Help tab, with your device details included.
Developers
The source code for both apps is on GitHub. Issues and suggestions are welcome.
SoundVisualizerFOR ANDROID
Phone appAndroid 10 or later
Draws the sound playing on your phone, from videos, music, games and more, around the edge of the screen, and gives each sound type its own vibration. Turn on External sound mode and it also listens to the sounds around you through the microphone.
Without leaving your video, music or game, it shows the direction and type of each sound with color, motion and vibration
Five scenes drawn around the edge of a phone screen. The color changes with the sound type, and the edge on the side the sound comes from is drawn larger.
Ambient
Human sounds
Danger
← Left
Right →
Four visual modes
Choose Wave, Pad, Circle or Outline. The graphics float transparently over other apps, and touches pass straight through to the app below. Size, Opacity, Sensitivity, Speed and Glow are saved separately for each mode.
Left and right
It compares the left and right channels of the sound playing on your phone, and draws which side the sound comes from and how loud it is. Phone audio has only two channels, so it can't tell front from back.
AI sound classification
Sorts sounds into Ambient, Human sounds and Danger, and shows each in its own color. Warning sounds such as sirens, alarms, explosions and horns count as Danger more readily, so they aren't missed even under music or game audio.
Choose the type for each sound
The 521 sounds the AI recognizes are shown in groups of similar sounds. For each group (or each sound, in Advanced mode) you choose which type it counts as. Your changes apply to the AI right away.
Vibration by sound type
For each sound type, choose a pattern (Off, Slow, Medium, Fast or Continuous) and a strength from 10 to 100%. At first only Danger is on, set to ‘Medium’. It vibrates while the sound lasts and stops when the sound stops.
Lock screen alert
If a Danger sound is heard while your phone is locked, an alert appears on the lock screen and the screen turns on for a moment. The alert makes no sound.
External sound mode
Instead of the phone's own sound, it listens to the sounds around you through the microphone and alerts you with the same graphics and vibration. Use it to notice fire alarms, sirens or someone calling you. Microphone sensitivity has 16 steps. This mode doesn't show direction.
Control without leaving your screen
From the running notification, stop the visualizer or switch the visual mode. With the Quick Settings tile, turn it on and off without opening the app.
Tutorial, help and 18 languages
The first time you open it, a tutorial shows how it works with the real drawing engine and real vibration. You can report problems straight from the Help tab. It supports 18 languages.
Try it yourself
Select a sound and tap the screen. The sound is drawn the same way SoundVisualizer draws it
Ambient
Overlay off: captions only
…
Regular captions
Choose a sound
Select a sound, then tap the spot on the screen where it should come from. Hold and move to drag it around.
AmbientHuman soundsDanger
Overlay settings
Visual mode
Output channels
Size 55
How far the wave can reach into the screen
Opacity 50
Higher values look more solid
Sensitivity 50
How fast it follows changes in volume. Low is gentle; high reacts instantly
How to use the demo
Real sound will play. Turn up the volume or use earphones.
Select a sound.
Tap where the sound should come from on the screen. Hold and drag to move it.
Change the settings above and compare the results.
Press Auto demo to step through changes in the direction, loudness and type of sound.
The real apps capture and analyze the sound playing on your phone or computer.
Direction and loudness come from comparing the volume of each channel, and a YAMNet-based AI identifies the type of sound.
A browser can't capture other apps' sound, so in this demo you set where the sound comes from yourself.
7.1 surround is available in the Windows version; the Android version tells left from right in the phone's sound.
The demo's colors are chosen to match this page; in the apps you set the color for each type yourself.
Which sounds count as Danger can also be changed for each sound in the app's Sounds tab, so for this wildlife documentary scene the demo puts the bear growl under Danger.
This demo is a trial version that runs in your browser.
Because a browser can't capture other apps' sound, it was built so that you set where the sound comes from.
The shapes on screen use the same drawing method as the apps (perimeter mapping, Catmull-Rom interpolation, EvenOdd shapes, channel smoothing),
while the confidence values on the badge are scripted. In the apps, the AI on your device decides the type of sound.
The photo in the demo is used under the Unsplash License.
The bear growl sound is Bear growl by Shizhao
(Wikimedia Commons, CC BY 3.0), trimmed to 1.75 seconds with the volume adjusted.
Install and turn it on
Install it from the APK file on a phone with Android 10 or later.
On your phone, go to GitHub Releases, then download and open SoundVisualizer-v1.8.4.apk.
If you're asked whether to allow installing unknown apps, allow it and install. On Galaxy phones (One UI 6 or later), turn off Auto Blocker for a moment, install, then turn it back on.
When you open the app, the tutorial comes first. On Home, tap Start and allow the display over other apps, microphone and screen recording permissions.
Open any app that plays sound, such as a video, music or game app, and the sound is drawn around the edge of the screen. To turn it off, tap Stop on Home, or Stop in the notification.
To listen to the sounds around you, turn on the External sound mode switch on Home before you start. It can't be changed while running.
Why does it ask for screen recording and microphone permissions?
Android requires these two permissions before it hands the sound playing on your phone to another app. Only the sound is used; the screen is never captured, and it doesn't listen to the sounds around you unless External sound mode is on. Under Android's policy, it asks for screen recording consent every time you start. On Android 14 or later, if you're asked what to share, choose Entire screen.
Which apps does it work with?
Most apps that play sound, such as YouTube, streaming services, music and games. A few apps block audio capture and can't be drawn (the app tells you when this happens), and call audio, notification sounds and alarms aren't captured. Also, nothing can be captured when the media volume is at 0, so to use it without sound playing out loud, connect earphones.
Can it tell me about sounds around me?
Turn on External sound mode and it listens through the microphone, alerting you with the same graphics and vibration. This mode doesn't show direction, and it can't hear anything during a call or while another app is using the microphone. What it hears is never saved to a file or sent anywhere.
Does it alert me when the screen is off?
When it shows your phone's own sound, by default it rests while the screen is off. Turn off Pause while the screen is off under Battery in Settings, and it will alert you with vibration even with the screen off. External sound mode keeps listening with the screen off regardless of this setting, and uses more battery as a result.
It turned off when I started a screen recording.
Screen recording, screen sharing and casting all use the same screen recording consent, and Android lets only one app use it at a time. So when another recording starts, the visualizer turns off, the phone gives three long vibrations, and a notification with a Turn on again button appears.
How is it different from the Windows version?
Choose the version that fits your device
Comparison of the two versions
Feature
Android
Windows
Sound source
Sound playing on the phone, or sounds around you through the microphone
All sound playing on the computer
Direction
Left and right (sounds around you appear in the center, without direction)
Stereo: left and right. 5.1 and 7.1: front, back and sides too
Sound types
Ambient, Human sounds and Danger. You can change which type each sound counts as in the Sounds tab
Visual modes
Wave, Pad, Circle and Outline. Size, Opacity and Sensitivity are saved separately for each mode
Vibration
Pattern and strength for each sound type
None
Turning on and off
App, running notification, Quick Settings tile
Launcher, tray icon
Languages
18 languages, including Korean, English, Japanese and Chinese
Captures all the sound playing on your computer and draws its direction on the screen, from stereo up to 7.1 channels. No installation: unzip it and run.
It draws every sound playing on your computer, from videos, music and games to video calls. With 7.1-channel output, it also tells front, back and sides apart
Five scenes drawn along the edges of a computer screen. The color changes with the sound type, and the edge on the side the sound comes from is drawn larger.
Ambient
Human sounds
Danger
← Left
Right →
Four visual modes
Choose Wave, Pad, Circle or Outline. Whatever maximum frame rate you set, from 30 FPS up to your monitor's refresh rate, the graphics move at the same speed.
Direction up to 7.1 channels
Detects stereo, 5.1 and 7.1 output automatically and draws each channel in its place. When the output device changes, it adjusts to that device, and you can also change it yourself.
AI sound classification
Sorts sounds into Ambient, Human sounds and Danger, and shows each in its own color. With 5.1 or 7.1, the AI listens to the rear and side channels too, not just the front.
Hotkeys without leaving your screen
Without minimizing a full-screen video or game, press F2 to switch the sound mode and F3 to switch the visual mode. You can change the hotkeys to any key combination in the Settings tab.
Edit mode
Press F4 and boundary lines appear on the screen. Drag them to set the size and limits of the drawing area. Settings such as color and opacity can be changed right in the panel.
Choose the type for each sound
The 521 sounds the AI recognizes are shown in the same groups as in the Android version. For each group (or each sound, in Advanced mode) you choose which type it counts as.
Tray icon
From the icon in the notification area, turn the visualizer on or off, open the launcher or exit the program.
Keyboard and screen readers
Every control in the launcher can be reached with the Tab key. Ctrl + Tab moves between tabs, and Esc closes windows. Screen readers such as Narrator and NVDA read out buttons and their states.
Tutorial, help and 18 languages
The first time you open it, a tutorial shows how it works with the real drawing engine. You can report problems straight from the Help tab. It supports 18 languages and starts in your Windows display language.
Run it without installing
A portable package for Windows 10/11 (64-bit). You don't need to install .NET separately.
Download SoundVisualizer_v2.1.1.zip from GitHub Releases and unzip it anywhere you like.
Run SoundVisualizer.exe in the folder.
If a ‘Windows protected your PC’ window appears, click More info, then Run anyway.
In the launcher, choose your language and finish the first setup, then click Start to turn on the overlay.
Settings are saved in settings.json next to SoundVisualizer.exe, so you can move or copy the whole folder.
With 5.1 or 7.1-channel output, it tells front, back and sides apart. Stereo output has only two channels, so it only tells left from right; sound that is on both sides is drawn at the top of the screen.
I can't see the graphics in a full-screen app.
If the app runs in exclusive full-screen mode, as some games do, try switching it to borderless windowed or windowed mode. Also check that the sound is playing through the Windows default output device and isn't muted.
The graphics barely react.
If an app's volume is turned very low or muted, in the app itself or in the Windows volume mixer, its sound may barely show. Turn that sound up a little or raise Sensitivity in the Settings tab.
Can I use it with a keyboard or a screen reader?
Yes. Every control in the launcher can be reached with the Tab key. In the tab bar, use the arrow keys; anywhere, Ctrl + Tab goes to the next tab. Dialogs and the overlay edit panel close with Esc. Screen readers such as Narrator and NVDA read icon buttons, hotkey buttons and the sound list by name, and every value you drag with the mouse in the edit screen can also be changed with the sliders in the panel.
Do I need to install it?
No. Unzip it and run SoundVisualizer.exe. You don't need to install .NET separately. Settings are saved in settings.json next to the program, so you can move the whole folder and keep using it.
How is it different from the Android version?
Choose the version that fits your device
Comparison of the two versions
Feature
Android
Windows
Sound source
Sound playing on the phone, or sounds around you through the microphone
All sound playing on the computer
Direction
Left and right (sounds around you appear in the center, without direction)
Stereo: left and right. 5.1 and 7.1: front, back and sides too
Sound types
Ambient, Human sounds and Danger. You can change which type each sound counts as in the Sounds tab
Visual modes
Wave, Pad, Circle and Outline. Size, Opacity and Sensitivity are saved separately for each mode
Vibration
Pattern and strength for each sound type
None
Turning on and off
App, running notification, Quick Settings tile
Launcher, tray icon
Languages
18 languages, including Korean, English, Japanese and Chinese
The sound playing on your device goes through four steps before it is drawn on the screen. Every calculation happens on the device.
How it works
No extra equipment, and no help needed from content creators
01
Capture the sound that's playing
It takes in the sound playing on your phone or computer as it is. There's no need to connect each app separately, whether it's YouTube, a streaming service, music or a game. On Android, External sound mode takes in the sounds around you through the microphone.
02
Work out direction and loudness
It measures and compares the volume of each channel to work out direction and loudness. Windows detects 2.0, 5.1 and 7.1 channels by itself; Android tells left from right.
03
AI sound classification
Every 0.25 seconds, the last second or so of sound is turned into a 16 kHz mono log-mel spectrogram, and a YAMNet-based ONNX model sorts it into Ambient, Human sounds or Danger.
04
Draw it on the screen
Loudness becomes size, direction becomes position, and type becomes color. What you're watching stays as it is, and Android also alerts you with vibration.
Four visual modes that keep your screen clear
The four screens below are being drawn live right now. Size, Opacity and Sensitivity can be set separately for each mode
Wave
Fills in along the edges of the screen, reaching further for louder sounds. You can read direction and loudness together, which suits watching videos.
Pad
Shows sound with pads fixed at eight points around the edge. It covers less of the screen while showing direction clearly.
Circle
A ring in the middle of the screen swells toward the side the sound comes from. You can read direction while keeping your eyes on the center.
Outline
Keeps only the outline of the wave. Of the four modes, it covers the least of the screen.
Design principles
An app that handles sound needs extra care
AI
We assume the AI can be wrong
Sound type detection isn't always right. So direction and loudness are calculated separately from the AI and always shown, and sounds you must not miss, like sirens and alarms, are caught as Danger more readily. When safety is at stake, don't rely on this app as your only safeguard.
Screen
What you're watching stays in view
The graphics float transparently over other apps, mostly along the edges of the screen. You decide how much of the screen they cover by changing the mode, size and opacity.
Choice
You decide the sound types
The 521 sounds the AI recognizes are shown in groups of similar sounds, and for each group you decide whether it counts as Ambient, Human sounds or Danger. You can change the colors too, and on Android, the vibration.
Privacy
Sound never leaves your device
Measuring loudness and telling sound types apart all happens on your device. Sound is never saved to a file or sent to a server, and it works without the internet.
Research
Tested with the people it's for
20people in the first test
2 roundswith the same students
4 of 4 itemspositive from every student in round 2
User tests at Seoul National School for the Deaf
In May 2026 we tested the Windows version twice. After making changes based on what we heard in round 1, we checked again with the same students
Round 1 2026.5.19–20
6 deaf and hard of hearing students and 14 hearing participants
They tried it themselves, then answered a 5-point survey. The information they wanted most was the direction of sound; their biggest worry was drawing their eyes away from what they were watching.
Changes
We made changes based on what we heard
We made the AI's decisions steadier and added more game sound effect data. We tuned how it responds to match loudness, and widened the range of settings. We also changed the defaults to a screen that stays comfortable over long viewing.
Round 2 2026.5.29
The same 6 students rated it again
Every student answered positively on all four items, and the biggest rise was in willingness to actually use it. After the test, the Windows version was installed in the school's computer lab.
Share of positive answers
Round 1 · all 20 participantsRound 2 · 6 deaf and hard of hearing students
The overlay is appropriate
1st88.9%
2nd100%
Satisfied with the direction display
1st91.7%
2nd100%
The AI colors are useful
1st72.2%
2nd100%
Would actually use it
1st66.7%
2nd100%
Round 1 shows answers from all 20 participants; round 2, from the same 6 deaf and hard of hearing students.
Field trial
In October 2026, we ran a field trial of the Android app with 6 deaf and hard of hearing students
Field trial 2026.10.21
6 deaf and hard of hearing students used the app themselves
At Seoul National School for the Deaf, they tried the Android app with the phone's own sound and in External sound mode, then answered a survey. The information they said they needed most was which direction a sound comes from.
Interview 2026.10.21
Advice from a sign language interpretation professional
We were advised to make the settings simpler, with pictures instead of text.
Applied 2026.10.22
Built into v1.8.0
We made the Simple settings view, where you choose by looking at pictures. Coughs and sneezes now also count as Human sounds, and these are shown sooner.
Share of positive answers
6 deaf and hard of hearing students · Android app
The loudness display is easy to understand
Oct100%
It doesn't get much in the way of using the screen
Oct83.3%
Vibration alerts are helpful
Oct83.3%
Would actually use the service
Oct83.3%
Would recommend it to others
Oct83.3%
“Just having it shown visually was a big help,” one student said. We're looking for ways to improve the lowest-rated item, the left–right direction display (3 of 6).
We're looking for institutions and organizations to join our next field trial.
AwardHonorable Mention at the 2026 Disability HackathonSoundVisualizer received an Honorable Mention in the finals of the Digital Inclusion track at the 2026 Disability Hackathon, hosted by the Korea Foundation for Persons with Disabilities.
ContestMid-stage build submitted to the Disability HackathonWe submitted a mid-stage build of SoundVisualizer for Android to the 2026 Disability Hackathon.
SelectedFinalist in the 2026 Disability HackathonSelected as one of 6 finalist teams in the Digital Inclusion track of the 2026 Disability Hackathon, hosted by the Korea Foundation for Persons with Disabilities.