Audio test

Five checks in order — left, right, stereo, microphone and latency — and a verdict that says what is broken, not just what you heard.

You should hear a tone from the left speaker or earpiece only. Turn the volume up a little if you need to.

Generated right here, in your browser. What you type never leaves your device.

Keep going with Speaker test Microphone test Audio latency test Morse code translator

Most audio tests answer a smaller question than the one you arrived with. A tone plays, you either hear it or you do not, and then you are on your own. That is fine when the fault is obvious and useless when it is not — because "no sound" has half a dozen causes and they need different fixes, and knowing which one you have is the entire problem.

This runs five checks in a fixed order and then says what the pattern means. Left channel, right channel, a tone that travels between them, the microphone with a real recording played back, and the round trip through the air that reveals how far behind your ears the sound actually is.

The order is the point. A microphone that seems dead when the speakers are also dead is not a microphone fault, and a stereo sweep that fails while both channels work individually means something quite specific. Reading the checks together is what turns five yes-or-no answers into a diagnosis.

Manual

How to use it

1
Put your headphones on, or turn the speakers up

The channel checks need you to be able to tell left from right, which is easier on headphones. On laptop speakers, sit centred, because a lot of laptops place both drivers close enough that the sides blur.
2
Answer honestly on each side

A tone plays on the left only, then the right only. If the second one sounds like it came from the same place as the first, that is a no — that answer is what separates a swapped cable from a dead channel.
3
Listen for travel, not for sound

The stereo step sweeps a tone from one side to the other. Both channels working but no sense of travel means the output is being folded to mono somewhere, which is a setting rather than a fault.
4
Record yourself and play it back

Watching a level meter twitch proves the microphone is connected. It does not prove the recording is usable. Three seconds played back tells you about hum, clipping, a fan, or a microphone pointing the wrong way.
5
Measure the round trip

Keep the microphone near the speaker and turn the volume up. Short bursts are played and timed as they come back through the air, which measures the whole path — including the Bluetooth delay the browser cannot see.
6
Read the verdict, not the score

The panel at the bottom names the fault it thinks you have and what to do about it. If a step does not apply, skip it rather than guessing; a skipped step is not held against the result.

In depth

Things worth keeping in mind

Reading the result patterns

The five answers combine into a small number of cases, and each points somewhere different. This is the table the verdict panel is applying.

Neither channel: almost never the speaker. Check which output device the system has selected — a forgotten Bluetooth speaker in another room is the classic.
One channel only: the plug, the cable, or the balance slider, in that order of likelihood.
Both channels but no travel: the output is being folded to mono. Both Windows and macOS have a mono-audio accessibility switch that does exactly this.
Speakers fine, microphone dead: browser permission first, then the selected input device, then a physical mute switch on the headset.
Everything works but latency is high: Bluetooth. Plugging in a cable will collapse the number, which is the quickest way to confirm it.

Why the latency number moves

A round trip through the air is a real measurement, but it is measuring your whole setup at that moment, and several things in that setup are not constant.

The codec is negotiated per connection. The same headphones can land on a different codec after a reconnect, which changes the delay.
Distance adds about 3 ms per metre. Sound travels roughly 343 m/s, so a microphone a metre from the speaker adds three milliseconds before anything electronic happens.
Other applications matter. A conferencing app holding the audio device open can force a larger buffer for everything else.
The browser itself adds some. A native application will usually measure lower than a web page on the same hardware.
Take the number as a band, not a figure. The useful question is which of the four bands you are in, not whether it says 186 or 194.

History

How sound got two sides

Stereo was patented in 1931 by Alan Blumlein, an engineer at EMI, reportedly after being irritated in a cinema that the actors' voices came from a single speaker regardless of where they stood on screen. His patent covers most of what stereo still is: two microphones at an angle, a way of cutting two channels into one record groove, and the psychoacoustics of why a level difference between two speakers makes a sound seem to come from between them. He filed it at twenty-eight and died in a test flight crash in 1942, aged thirty-eight.

The record industry took another twenty-five years to use it. Stereo LPs arrived in 1958, using the vertical-lateral groove system Blumlein had described, and for years many releases were issued in both mono and stereo because most people owned one speaker. Some famous records were mixed carefully in mono and hastily in stereo, which is why the mono versions are the ones enthusiasts hunt for.

Latency became a problem consumers could feel much later, when audio stopped being a wire. An analogue cable delays sound by the time light takes to cross it — nanoseconds. Bluetooth has to compress the audio, packetise it, transmit it, buffer against interference, decompress it and play it, and the standard SBC codec commonly lands somewhere between 150 and 250 milliseconds. That is a quarter of a second, which is unmissable on video and unusable for playing an instrument.

Which is why the codecs became a selling point. aptX Low Latency, LDAC, LC3 in Bluetooth LE Audio: all of them are attempts to claw back the delay that the wire never had. The number that matters is the one measured through the air on your own hardware, because it depends on the codec both ends negotiated, not on the one printed on the box.

1931: Blumlein patents stereo
1958: the first stereo LPs on sale
~200 ms: typical Bluetooth delay on the SBC codec
40 ms: roughly where a lag stops being noticeable

How it works

What each check can actually prove

The channel tests do not pan a tone, they build it on one side. Panning is a level difference between two channels, and a system set to mono will happily average a hard-panned tone back into the middle — so a panned test would pass on a machine that has folded your stereo away. Generating the tone with a stereo panner set hard over and checking whether travel is audible is what catches that case.

The microphone check is deliberately two-stage. A level meter reads the peak of a short block of samples, which answers "is anything arriving" and nothing else. It cannot tell you the gain is so high everything clips, that the automatic gain control is pumping, or that you are speaking into the wrong edge of a laptop. Playing three seconds back is the only check that puts your ears on the actual signal.

The meter is drawn in decibels rather than linearly. Perceived loudness is roughly logarithmic, so a linear meter spends nine tenths of its travel on the loudest tenth of what you can hear, and a normal speaking voice barely moves it. Half amplitude is −6 dB, which sits most of the way up a decibel meter and exactly halfway up a linear one; the first is the one that matches what you are hearing.

Latency is measured by sending sound out and listening for it to come back. The browser will tell you its own `outputLatency`, but that number describes the software path only — it stops at the point the operating system hands the audio over, and a Bluetooth headset adds its delay well after that. A round trip through the air is the only measurement here that sees the whole chain, which is exactly why it is worth doing and why no other page on this search result offers it.

The bursts are 2 kHz rather than clicks. A click has its energy spread across the whole spectrum, and small speakers reproduce almost none of the part that matters, so the onset is hard to detect. A short tone burst puts the energy where laptop speakers and phone drivers are actually efficient, and the median of five trials throws out the one where a door closed.

By hand

Finding a channel fault without any tool

Worth knowing because it isolates the fault to the cable, the socket, the device or the software in about a minute, and none of it needs a website.

1

A familiar track, not a test tone. You will notice a missing instrument long before you notice a missing frequency band.
2

If sound cuts in and out or one side comes and goes as you move it, the fault is the connector or the cable, and nothing in software will fix it. A plug that is not pushed fully in is the single most common cause of one-sided sound.
3

Put the left cup on your right ear. If the silence moves with the cup, the fault is in the headphones. If it stays on the same side of your head, the fault is upstream — the socket, the device, or a balance setting.
4

Different device, same headphones. If both sides work there, the headphones are fine and you have narrowed it to the computer.
5

Every operating system has one and it is nearly always where you are not looking. A balance dragged fully to one side is indistinguishable from a dead channel until you find the slider.

Diagnosis

If something is not working

Symptom:The first tone plays but I cannot tell which side it came from

Likely cause:: Laptop speakers often sit close together, and some are angled downwards, which blurs the stereo image at close range.
What to do:: Use headphones for the channel steps. If you have none, cup a hand behind each ear in turn — whichever side gets louder is the side that is playing.

Symptom:The browser never asks for microphone permission

Likely cause:: Permission was denied earlier and the browser remembered, so it now refuses silently rather than asking again.
What to do:: Click the padlock or the sliders icon in the address bar, find the microphone entry, set it back to ask or allow, and reload the page.

Symptom:The level meter moves but the recording is silent

Likely cause:: A different input is selected than the one you are speaking into, and the one being metered is picking up room noise.
What to do:: Open the system sound settings, pick the input you mean, and test again. This is exactly the case the playback step exists to catch.

Symptom:The latency measurement gives up

Likely cause:: The microphone is not hearing the bursts, usually because the volume is too low or echo cancellation is removing them on purpose.
What to do:: Turn the volume up, hold the microphone closer to the speaker, and use a quiet room. On headphones, hold one earpiece against the microphone.

Symptom:Latency reads over 200 ms and I am on a cable

Likely cause:: Something else has the audio device open with a large buffer, or the browser is running the audio through a virtual device.
What to do:: Close conferencing and recording applications, then measure again. If it persists, check whether the output device is a virtual mixer rather than the hardware.

Symptom:The verdict says the test is incomplete

Likely cause:: The two channel steps have not both been answered. Everything else is read against them, so nothing can be concluded until they are.
What to do:: Answer left and right, even if the answer is no. A negative answer is what produces the most useful diagnosis.

Trivia

Things worth knowing

Your ears resolve ten microseconds

Human hearing locates sounds partly by the difference in arrival time between the two ears, and that mechanism is sensitive to around 10 microseconds — a hundred-thousandth of a second. It is why a stereo image collapses so obviously when one channel is delayed, and why a badly synchronised pair of wireless earbuds sounds wrong rather than merely late.

The mono switch is an accessibility feature

Both Windows and macOS can fold stereo to mono, and it is there for people with hearing in one ear, who would otherwise miss whatever was mixed to the other side. It is also switched on by accident constantly, and it is the usual reason both channels work but nothing travels between them.

Bluetooth audio is never uncompressed

Even the highest-quality Bluetooth codecs compress, because the radio link does not have the bandwidth for CD-rate audio. SBC, the mandatory baseline every device supports, is the fallback when two devices cannot agree on anything better — and the one most likely to be running when the latency measurement comes back high.

Broadcasters have a legal tolerance for lip-sync

The ITU recommends audio be no more than about 90 ms behind video or 185 ms ahead, and several regulators enforce something similar. Audio ahead of picture is tolerated less, because it never happens in nature: sound always arrives after the sight of what made it.

A 2 kHz burst is easier to detect than a click

A click contains every frequency at once, which sounds like it should be easy to spot. In practice most of its energy sits where small speakers are inefficient, so what reaches the microphone is a smeared thud. A short tone burst concentrates the energy where laptop and phone drivers are actually loud.

Questions

Frequently asked questions

Do I need a microphone for this?

Only for the last two steps. The three speaker checks work with no permissions at all, and the microphone and latency steps can be skipped. A skipped step is not counted as a failure.

Does my recording get uploaded?

No. The three seconds are held in the page and played back from memory, and they are discarded when you leave. There is no server here to send audio to.

Why measure latency through the air instead of asking the browser?

Because the browser only knows about its own software path. It has no idea a Bluetooth headset is adding two hundred milliseconds after the operating system has handed the audio over. Playing a sound and listening for it is the only way to see the whole chain.

What is a normal latency figure?

Under 40 ms is wired territory and imperceptible. Between 40 and 100 is normal for decent wireless. Past 100 you will see lip-sync drift on video, and past 250 it is obvious to everyone. The band matters more than the exact figure.

Both channels play but the sweep does not travel. What is that?

Almost always a mono setting. Both Windows and macOS have an accessibility switch that folds stereo into a single signal sent to both sides, and it is easy to enable by accident. Less often, the channels are wired the same way at the source.

How is this different from your speaker test and microphone test?

Those go deep on one device each — frequency sweeps, channel identification, spectrum. This one runs the short version of several in sequence and then interprets the pattern. Start here to find out what is wrong, then use the specific tool to dig in.

Can I use it on a phone?

Yes. The channel steps need headphones on a phone, because the built-in speakers are too close together to separate. Everything else works the same.

Why did my latency change between two runs?

Bluetooth codecs are negotiated per connection, so reconnecting can land you on a different one. Distance from the speaker adds about 3 ms per metre, and another application holding the audio device can enlarge the buffer for everyone.