You say "hey, how's it going?". The stranger on the other side opens their mouth, then freezes: they've just heard their own voice come back, a quarter of a second late, with a faint whistle behind it. They frown, mouth "you've got an echo", you answer "what?", and they hit next. You, in your headphones, heard nothing unusual — and that's exactly the problem: the echo only exists in the other person's ear.
We've already covered picture and sound in video dating, mastering your webcam before you even connect and the voice that matters more than the face. This article tackles the most common and most badly diagnosed audio flaw in random video chat: acoustic echo, and its violent cousin, feedback. It's a peculiar fault because it's asymmetric: you never hear it, the person opposite hears far too much of it. And since nobody on chatroulette takes the time to explain, you rack up "nexts" without ever understanding why.
Three parts: exactly how echo works, a five-minute diagnosis, then the fixes, from free to hardware — with the precise system settings to check.
1. Where the echo actually comes from
A loop between your speaker and your mic
The principle is mechanically simple. The other person speaks. Their voice comes out of your speakers. Your mic, sitting a few centimetres from those speakers, picks up that voice along with your own. And it sends it back out over the network. The result: they hear themselves, delayed by the network round trip — typically 150 to 400 milliseconds.
That delay is the key. Anything under 30 ms is perceived as a mere colouring of the voice, unpleasant but bearable. Beyond 100 ms, the brain identifies it as a separate sound, and that's when talking becomes physically painful: you trip over your own voice, lose your train of thought, slow down. Research on delayed voice feedback (the phenomenon known as delayed auditory feedback, documented since the 1950s) shows that a lag of this magnitude is enough to cause stuttering and involuntary pauses in an otherwise normal speaker. The person opposite isn't leaving out of rudeness: they literally can no longer put sentences together.
Feedback: the loop that feeds itself
Feedback (or larsen) is the next stage. If both sides have open speakers, the sound goes round in circles: mic A → speaker B → mic B → speaker A → mic A… With each lap, the frequencies favoured by the room and the equipment gain in intensity. Within two or three seconds, you get the high-pitched squeal everyone knows. It's unpleasant, and on headphones turned up loud, it can actually hurt your ears. It's also why feedback leads to an instant disconnect, no discussion.
Why automatic cancellation isn't always enough
Modern browsers ship with software echo cancellation. In the WebRTC specification used by virtually every in-browser video chat platform, it's called echoCancellation and it is enabled by default on captured audio streams — Mozilla's MDN Web Docs lists it among the standard getUserMedia constraints. The algorithm continuously compares what it sends to your speakers with what it receives from the mic, and subtracts the latter from the former.
It works very well, except in five cases:
| Situation | Why cancellation fails |
|---|---|
| Speaker volume very high | The captured signal saturates the mic, subtraction becomes impossible |
| External mic + separate speakers | The algorithm doesn't know the acoustic path between the two devices |
| Very reverberant room (tiles, bare walls) | The echo "tail" exceeds the algorithm's analysis window |
| Mic grabbed by third-party software (OBS, virtual mixer) | The stream no longer passes through the browser pipeline |
| Badly connected headphones / sound still coming out of the speakers | The user believes they're on headphones, they're not |
That last case is by far the most common. A USB headset plugged in after the tab was opened isn't always picked up as the output by the page: sound keeps coming out of the laptop speakers even though the headset is on your head. You hear none of the echo, since you don't even hear the other person in the right device — but your mic picks up everything.
2. Diagnosing your own audio in five minutes
The bare recording test
You can't judge your own sound by talking to yourself. You have to record yourself under real conditions, meaning with the other person's voice coming out of your speakers.
The protocol, to do once and for all:
- Open any voice recording app (Voice Recorder on Windows, Voice Memos on macOS/iOS, Recorder on Android).
- Play any video on your computer, at the volume you normally use for chatting.
- Record fifteen seconds: five seconds of silence, five seconds speaking normally, five seconds of silence.
- Listen back on headphones, at moderate volume.
The verdict:
- You can hear the video during the silent stretches → your mic is picking up your speakers. Guaranteed echo for the other person.
- You only hear a steady hiss → a gain problem, not echo.
- You hear your voice clearly and clean silence → your audio is fine, the problem lies elsewhere.
The three other audio mistakes that get you skipped
Echo isn't alone. In order of how often it shows up in random video chat:
The mic too far away. A built-in mic 60 cm from your face picks up as much of the room as it does of you. Your voice arrives buried, the other person turns the volume up… which makes the echo worse. The classic vicious circle.
Gain too high. A mic cranked to maximum compresses and clips the moment you laugh. The other person gets a crack in the ears every time you burst out laughing. It's the most hated flaw on platforms where volume isn't normalised.
Contact noise. Mechanical keyboard, mouse, fingernails on the desk, a creaking chair: a mic sitting on the desktop transmits those vibrations through the furniture, not through the air. Noise processing almost never filters them out.
To those three, add an underrated classic: the fan. A laptop heating up during a video call runs at full tilt, and its whoosh is ten centimetres from the built-in mic. A simple raised laptop stand that clears the vent grille cuts both the temperature and the captured noise — a double win for a few euros.
3. Fixing it: from free to hardware
Level 0 — free, instant, effective in 80% of cases
Put on headphones. That's it. If the other person's sound no longer plays into the room, your mic can't pick it up, and the echo disappears by design. No algorithm beats physical separation.
But you still have to check that the sound actually comes out of the headphones:
- Windows: Settings → System → Sound. Make sure "Output" really shows your headset, then go down to "Advanced sound settings" → "Volume mixer" and check that the browser is using that same device. Windows allows different outputs per application: a frequent trap.
- macOS: System Settings → Sound → Output. Then, in the browser, reload the tab after plugging in the headset.
- Android / iOS: plug in or pair the headset before opening the chat page. Switching to Bluetooth mid-call sometimes kills the mic.
- In the browser: click the padlock or camera icon in the address bar to see and change the mic the site is using.
If you use in-ear headphones, go for a wired model: Bluetooth adds 100 to 200 ms of latency, which stacks on top of the network delay and spoils conversational comfort. For video chat, a wired headset with built-in mic remains the best value for money on the market.
Level 1 — gain settings
Second move, three minutes:
| Setting | Recommended starting value |
|---|---|
| Mic volume (Windows / macOS) | 60 to 75%, not 100% |
| Mic boost ("Microphone Boost", Windows) | 0 dB, raise only as a last resort |
| Speaker volume if no headphones | The lowest that's still audible |
| Mouth-to-mic distance | 15 to 25 cm |
The universal mistake is cranking the mic to maximum because "people can't hear me". A mic at 100% picks up the whole room; a mic at 70% placed closer picks up your voice. Move closer before raising the gain: the difference is roughly 6 dB every time you halve the distance.
Level 2 — the hardware that genuinely changes things
If you do video chat regularly, three modest purchases cover 95% of needs, in this order of priority:
- A closed-back headset with a mic. The closed earcup prevents leakage into the mic, and the boom or arm puts the capsule near your mouth. This is investment number 1, way ahead of the webcam.
- A clip-on lavalier mic if you'd rather keep your head bare on camera: discreet, sitting 20 cm from your mouth, it sorts out distance and gain in one go. A wired USB-C lavalier mic does the job nicely without venturing into the world of audio interfaces.
- A boom arm or a decoupled stand if you use a desk-mounted mic. A suspended arm eliminates vibrations transmitted through the desk — keyboard clacks vanish.
And for reverb, there's no need to line the whole room: a thick curtain, a shelf full of books, a rug and a bed already do an excellent job. If your video corner is tiled or faces a bare wall, a few acoustic foam panels behind the screen are enough to kill the early reflections — the ones that give you that bathroom sound echo cancellation hates.
Level 3 — software processing, with caution
Machine-learning noise suppressors, built into some sound card drivers or available as dedicated software, work wonders on steady noise (fan, air conditioning, street). But three caveats:
- They damage soft voices: the detection threshold cuts off the ends of sentences and whispers.
- They eat CPU, which can degrade the video stream running alongside.
- They never replace physical separation between speaker and mic.
Simple rule: sort out the physics first (headphones, distance, gain), software last, and never two noise processors stacked on top of each other — the browser's and the driver's fight each other and produce a sound that "pumps".
4. The social reflex: tell the other person
One last point, and it's a matter of etiquette. Since you never hear your own echo, you depend entirely on others to know about it. So return the favour:
- If someone has echo, say it in one short, neutral sentence: "you've got an echo, try headphones". No blame, no diagnosis.
- Write it in the text chat too: if there's feedback, the audio is unusable, but text gets through.
- If someone points it out to you, don't argue: plug in headphones, it's fixed in five seconds. Plenty of people waste thirty seconds explaining that "that's impossible, I can't hear anything" — of course you can't, you're physically unable to hear it.
- And if the echo persists while you're on headphones, check that your browser output really is the headset, not the speakers.
Key takeaways
| Symptom for the other person | Most likely cause | Fix |
|---|---|---|
| They hear themselves on a delay | Mic picking up your speakers | Headphones |
| High-pitched squeal | Feedback, both sides without headphones | Lower the volume, headphones |
| "I can barely hear you" | Mic too far away | Move closer before raising the gain |
| Cracks when laughing | Gain clipping | Mic at 70%, boost at 0 |
| Constant hiss | Fan or gain too high | Raise the laptop, lower the gain |
Audio is the neglected part of video chat prep: people spend ten minutes adjusting a lamp and zero checking their mic. Yet in video dating, average picture quality is forgiven, painful sound never is. A fifteen-euro headset and five minutes of testing will win you more conversations than any accent light. And unlike lighting, nobody will see that you made the effort — which is precisely the sign that it worked.
