How Much Audio Delay Is Noticeable?

Under about 20 milliseconds, almost nobody notices audio delay. Between 40 and 50 ms it starts showing up as lip-sync drift on a talking face. Past roughly 100 ms it feels like a bad phone call. So how much audio delay is noticeable depends less on the number itself than on the job you are asking the audio to do.

The single biggest factor is whether you are listening to someone else or listening to yourself. Passive listening tolerates a lot. Hearing your own voice come back at you tolerates almost nothing. If you have already run the round-trip latency test and are holding a number, this page is how to read it.

The three thresholds worth knowing

Audio delay does not have one threshold. It has a staircase, and each step corresponds to a different perceptual mechanism failing.

These are working thresholds, not laws of physics. Individuals differ, trained listeners do better, and a quiet room makes everything more obvious. Treat them as the boundaries where complaints start, not as a pass/fail line.

How much audio delay is noticeable when you hear yourself

This is where the numbers get brutal. When you speak, your own voice reaches you twice: through the air and through bone conduction inside your skull, both essentially instantly. A delayed copy arriving from a speaker gets compared against that internal reference, and the comparison is far sharper than anything you can do watching a screen.

The effect has a name. Delayed auditory feedback is well enough documented that it is used deliberately as a speech-disruption technique: feed people their own voice back late enough and they stammer, slow down, or lose the thread of a sentence. You do not need laboratory equipment to reproduce it. Put wireless earbuds in, open a live monitor, and try to read a paragraph aloud.

The practical consequence: an audio latency threshold that is completely acceptable for playing music is not acceptable for a microphone you are talking into while the speaker is in earshot. If you can hear the speaker, you are in the strict regime. If the speaker is behind you facing a crowd, and you are mostly hearing your own unamplified voice, you are in the forgiving one.

Lip sync: the 40-50 ms line

Lip sync is the threshold most people have an intuition for, because everyone has watched a stream where the mouth and the words came apart. The commonly used lip sync delay threshold sits around 40 to 50 ms of audio lag before a face on screen looks wrong to an average viewer.

Two details matter. First, this applies to a visible face; a wide shot of a stage hides much more slop than a close-up. Second, audio arriving late is tolerated better than audio arriving early, because in the physical world sound always lags the sight of an event. Your brain has spent a lifetime accepting late sound from across a room and has no template at all for early sound.

That asymmetry is why Bluetooth delay is annoying rather than uncanny. It always errs in the direction the brain already expects, just far past the amount it expects.

What Bluetooth actually delivers against those numbers

Now put real link latencies next to the perceptual staircase. These are published figures, and the sources genuinely disagree with one another, so they are presented as ranges rather than averaged into a single tidy number.

Read that against the staircase and the conclusion is unavoidable: an ordinary iPhone-to-Bluetooth-speaker link lands in the "bad phone call" band, not the "invisible" one. That is a property of the codec and the receiver's buffer, not of any app. No app on your phone can remove Bluetooth latency, because the delay is not happening on your phone. The full argument, with the pipeline broken into stages, is in the guide on where Bluetooth microphone delay actually comes from, and the codec-by-codec detail is in the sourced comparison of SBC, AAC and aptX.

If you are shopping and specifically want to know whether the iPhone can reach the 40 ms tier, the short answer is in the page on iPhone and aptX Low Latency. It cannot, and no accessory changes that.

Measure yours before you judge it

Published codec figures describe one hop. What you actually experience is the whole round trip: microphone capture buffer, processing, encode, radio, the receiver's own buffer, and the speaker's converter. Speakers vary enormously in that last part, and a cheap one with a generous buffer can be worse than an expensive one.

The acoustic loopback latency test on this site measures the real number instead of guessing it. It plays a 30 ms chirp sweeping 1 to 6 kHz, listens on the microphone, cross-correlates the two signals to find the offset, runs five times, takes the median, and subtracts the flight time of sound through the air for the distance you enter. What comes back is your delay, on your hardware, in your room.

How much delay is too much for what you are doing

The same measurement means different things depending on the task. Roughly, from strictest to loosest:

• Singing with a backing track. Strictest case in normal life. You are locking your voice to a rhythm you can hear. Over about 100 ms this stops being a nuisance and starts being impossible. Sending the backing track down the same Bluetooth path does not fix the timing, it doubles it: you sing along to the track as it comes out of the speaker, so your voice then makes its own trip through the capture buffer, the encoder and the receiver's buffer, and arrives roughly one Bluetooth latency behind the music the room is hearing. The karaoke delay guide works through the arrangements that genuinely help.
• Live monitoring your own voice. Also strict, for the bone-conduction reason above. Aim for the low end or take the monitor out of your ears.
• Talking over video. Governed by the lip sync threshold. Fine at 40 ms, visibly wrong well before 100.
• Amplifying a voice to a room you are facing. Surprisingly forgiving. If you cannot hear the speaker over your own voice, a hundred-odd milliseconds mostly goes unnoticed by you, and the audience has no reference to compare against.
• Announcements, tour groups, a coach on a field. The most forgiving of all. Nobody is syncing to anything.

That last group is why phone-as-microphone setups work better than the raw figure suggests: nobody in the room has an undelayed copy to compare against.

What you can actually change

Once you have a number, the options are short and mostly physical.

1. Go wired. Bluetooth is not the route into the under-20 ms band. A cable gets you there, and so do dedicated 2.4 GHz wireless microphone systems, which are purpose-built radio links rather than Bluetooth ones; quoted figures vary by model, so check the current spec sheet for the one you are considering. The wired versus Bluetooth comparison puts the two side by side without pretending the convenience cost is zero.
2. Shorten the chain. One speaker, not two. Two Bluetooth speakers introduce their own problem, described in the guide on speakers drifting out of sync.
3. Move the speaker away from you. This does not reduce the delay, and it cannot remove your internal reference copy, since you always hear your own voice through the air and through bone conduction inside your own head. What it does is drop the level of the delayed copy reaching you, so the late voice stops competing with the live one, which makes it easier to keep talking. It also helps with feedback and howling.
4. Stop looking for a software fix. Apps advertising otherwise are worth a skeptical look, and that claim gets measured in the page on apps that promise the delay away.

The mic-to-speaker app this site is built around takes the honest position: it routes your iPhone's microphone into whatever output iOS is already using, at whatever latency that route has, and it tells you what the route is rather than promising a number it cannot deliver. Pairing still happens in iOS Settings, because that is the only place any iPhone app can leave it.

If your number came out worse than you expected, work through all the guides for your specific speaker and setup.