Speak into your mic and hear your own voice come back a beat later — that’s monitoring latency: the time your voice spends traveling mic → converter → OS → software → converter → headphones. Under ~10 ms it feels like normal sidetone; past ~30–40 ms it registers as echo and starts sabotaging your speech.

The pipeline that sets the delay

Digital audio isn’t processed continuously — it’s chopped into buffers: fixed-size chunks of samples handed between stages. Buffer size is the biggest lever you control:

• At the common 48 kHz rate, a 256-sample buffer is ~5.3 ms — but that’s one buffer on one leg. Input buffer + processing + output buffer + converter latency stack up; real round-trips through a desktop audio stack land anywhere from ~20 ms (small buffers, a decent driver) to 100 ms and beyond (large buffers, Bluetooth output).
• Smaller buffers = less delay, more risk. Fewer samples per chunk means the CPU must service audio more often; miss a deadline and you get crackles and dropouts, not elegance. 64–128 samples is the monitoring range; 256–512 is where recording setups live.
• The destination can be the bottleneck. Monitoring over Bluetooth adds its own codec delay on top — wired headphones only when you’re judging latency.

Buffer size lives in the interface’s driver panel or your recording app’s audio settings — browsers don’t expose it, which is part of why software monitoring inside a web app is a lost cause.

Where the echo is actually coming from

If you’re hearing yourself right now, one of these is on:

• Windows ‘Listen to this device’ — Sound Control Panel → Recording → your mic → Properties → Listen tab. The checkbox routes the mic to your output through the whole OS stack. It’s a diagnostic feature that people toggle while debugging, forget, and then live with a 50–100 ms echo. Uncheck it once you’re done testing.
• App monitoring — OBS’s monitoring device (Advanced Audio Properties), a DAW’s input monitoring, mixer/routing utilities, headset companion apps with a sidetone or monitor slider.
• Call-side echo — a different animal wearing the same name. If others hear themselves echoed, your speakers are feeding your mic. Headphones fix it outright; echo cancellation — a negotiated flag you can see in the tester’s info grid — is the software patch. That problem lives in the far end’s chain, not yours.

When monitoring is worth doing at all

• Calls and meetings: never monitor. You gain nothing — the far end already tells you if you’re audible — and delayed sidetone only distracts.
• Podcast, stream, voiceover: monitor, but in hardware. A USB mic with a headphone jack, or an interface with a Direct Monitor switch, splits the signal to your ears before the digital round trip — effectively zero latency, no CPU risk. It’s a real reason to buy that tier of gear; USB mic vs headset vs built-in covers the trade-offs.
• Music recording: same hardware rule, plus a low-latency monitoring path in the DAW and conservative buffer settings.

What the tester can and can’t tell you

The mic tester reads the capture side only: level, waveform, negotiated flags, channel count — the numbers that prove the mic and OS are healthy. It can’t measure monitoring latency (that lives on the playback side), and it won’t feed your voice to your speakers — by design, that’s a feedback generator in a browser tab.

What it can settle is which half of the problem you have. Meter moves, recording clean → capture is fine; if you still “hear echo,” the cause is a monitoring toggle or the far end’s speakers, not your mic. The playback checklist covers what to listen for once the clip exists.

Frequently asked questions

Why do I hear my own voice delayed in my headphones?

Somewhere in your chain the mic is being routed back to your ears through software: Windows' 'Listen to this device', an app's monitoring feature (OBS's monitor device, a DAW's input monitoring), or a mixer utility with a sidetone slider. Software monitoring always carries the full round-trip latency — often 50–100 ms — which your brain hears as echo. Find the toggle and turn it off; hardware direct monitoring exists precisely because this path is too slow.

What buffer size should I use for microphone monitoring?

For live monitoring, the smallest that doesn't crackle — typically 64 or 128 samples. For recording while monitoring through hardware, 256–512 is safe. Each doubling of the buffer roughly doubles that leg of latency; at 48 kHz, 256 samples is ~5.3 ms per buffer, and you have at least two buffers (input and output) plus converter time. The setting lives in your interface's driver panel or your recording app — browsers don't expose it.

Is monitoring delay the same as the echo people hear on calls?

No — two different problems with the same name. Monitoring delay is you hearing *yourself* late. Call echo is the *other* side hearing themselves back: your speakers bleed into your mic and their voice returns a beat later. Echo cancellation — the flag shown in the tester's info grid — exists for the second problem; headphones eliminate it mechanically.

Why doesn't this mic tester let me listen to myself live?

Deliberately. The capture chain is never connected to the speaker output — piping a live mic to local speakers inside a page invites feedback howl, and browser-pipeline latency would make it useless as monitoring anyway. The honest substitute is the 60-second clip: record, play back, judge.

When do I actually need zero-latency monitoring?

Podcasting, streaming, voiceover and music recording — any time you must hear your own voice while performing. A USB mic with a headphone jack, or an audio interface with a Direct Monitor switch, routes the analog signal to your ears before the digital round trip, so the delay is effectively zero. For meetings, don't monitor at all — it only adds distraction.

Why does a delayed copy of my voice make me stumble over words?

Delayed auditory feedback — your speech system expects near-instant confirmation of what it produced, and hearing it ~50+ ms late disrupts fluency (it's literally how 'speech jammer' toys work). If monitor delay makes you stumble, it's real, not imagination: switch to hardware direct monitoring or stop monitoring.