Milliseconds to Microseconds

Buffer Settings Are Displayed in Milliseconds and Built Out of Microseconds

An audio interface panel offers buffer sizes as sample counts — 64, 128, 256, 512 — and the DAW politely translates them into milliseconds underneath. Neither number is the one the hardware works with. The interface deals in single samples, each lasting a few tens of microseconds, and every latency figure in the session is just a count of those samples dressed up in a friendlier unit. Dropping the millisecond figure back to microseconds is how a buffer choice, a phase alignment and a mic-distance offset end up on the same scale.

Conversion factor: 1 ms = 1,000 µs, so multiply the millisecond figure by 1,000. A 128-sample buffer at 48 kHz is 2.6667 ms, which is 2,666.7 µs — exactly 128 sample periods of 20.8333 µs each.

The Numbers Behind the Slider

The sample period is the atom

At 44.1 kHz one sample lasts 22.6757 µs, at 48 kHz 20.8333 µs, at 96 kHz 10.4167 µs. Nothing in the signal path can be timed more finely than that, and every other figure is a multiple of it.

A buffer is a count, not a duration

Because the setting is in samples, raising the sample rate shortens the same buffer. Move 256 samples from 48 kHz to 96 kHz and the block halves from 5.3333 ms to 2.6667 ms, at the cost of twice the processing load.

Ten milliseconds is where players start noticing

Below roughly 10 ms of round trip most performers hear their monitoring as immediate; past 20 ms the delay becomes something to fight. That threshold is 10,000 µs, or about 480 samples at 48 kHz.

The block is paid for twice

One buffer fills on the way in and another empties on the way out, so the round trip already contains two of them before converters or driver offsets are counted.

Sizing a Buffer for the Take You Are About to Record

Tracking and mixing want opposite settings, and the decision is easier once the displayed millisecond figure is expressed in the samples and microseconds the interface is really counting.

Type the millisecond figure your DAW shows

Enter 2.6667, 5.3333, 11.61 — whatever the audio preferences panel reports for the current buffer. The microsecond value updates while you type, and a comma decimal separator is accepted as readily as a dot.

Divide by the sample period to get the block size

2,666.7 µs ÷ 20.8333 µs is 128 samples, which confirms the driver is reporting the buffer and nothing else. If the division lands on an odd number, the figure already includes a safety offset.

Swap when you start from microseconds

The swap control (↔) turns the pair into µs → ms, the direction you want for a delay-compensation offset, a mic-distance correction or a converter's group-delay spec quoted in microseconds.

Copy the value into a delay plugin

The copy control hands over digits only, so the number goes straight into a sample-delay or alignment field without a unit label to delete first.

The displayed figure is not the whole delay: converter group delay and the driver's safety offset sit outside the buffer, and plugins with lookahead add their own. A loopback measurement is the only way to learn what the session is really doing.

Buffer Size Against Sample Rate, One Block at a Time

Each row is one buffer setting; the middle columns give how long that block lasts at the three usual session rates, and the last column shows the 48 kHz figure in microseconds. These are single-direction block times — a round trip contains at least two of them.

Buffer (samples)	At 44.1 kHz	At 48 kHz	At 96 kHz	48 kHz block in µs
64	1.4512 ms	1.3333 ms	0.6667 ms	1,333.3 µs
128	2.9025 ms	2.6667 ms	1.3333 ms	2,666.7 µs
256	5.8050 ms	5.3333 ms	2.6667 ms	5,333.3 µs
512	11.6100 ms	10.6667 ms	5.3333 ms	10,666.7 µs
1024	23.2200 ms	21.3333 ms	10.6667 ms	21,333.3 µs

The pattern is worth memorising: doubling the buffer doubles the block time, doubling the sample rate halves it. That is why a 512-sample setting is fine for mixing, where nothing is being played in, and unusable for overdubbing at 44.1 kHz, where two of those blocks alone already exceed 23 ms before a single converter is counted.

Useful While the Session Is Open

Type the figure your DAW displays

Both boxes are live, so you can run down a list of buffer settings and see each one in microseconds without clearing the field between entries.

Turn it round for a sample period

Swapping to µs → ms suits alignment work, where the number starts as a handful of microseconds and has to be judged against a millisecond-scale monitoring budget.

Four decimal places of block time survive

Sample periods are recurring decimals, and output carries up to eight of them, so 20.8333 µs at 48 kHz stays exact enough to multiply back into a buffer count.

Take lengths are on the same dropdown

The searchable unit lists run from nanoseconds to minutes, so a five-minute take and a 20-microsecond sample period can be handled without opening a second page.

Questions From the Tracking Room

Which buffer setting gets me to about 5 ms?

At 48 kHz it is 256 samples, which gives 5.3333 ms — 5,333.3 µs. The same 256 samples at 44.1 kHz stretches to 5.8050 ms, and at 96 kHz shrinks to 2.6667 ms, so you would need 512 there to land near five. Bear in mind this is one direction only; if the goal is 5 ms of round trip rather than 5 ms of buffer, drop to 64 or 128 samples and let the converters take up the rest.

How long is a single sample at each session rate?

Divide one second by the rate: 22.6757 µs at 44.1 kHz, 20.8333 µs at 48 kHz, 10.4167 µs at 96 kHz, and 5.2083 µs at 192 kHz. It is a handy figure to keep in mind because it converts directly into distance — sound covers about 7.15 mm during one 48 kHz sample, so nudging a mic by a centimetre is worth roughly a sample and a half of alignment.

Why is my measured round-trip latency larger than the buffer figure?

Because at least four things stack up on top of it. The buffer is counted once on input and once on output. The analogue-to-digital and digital-to-analogue converters each add group delay from their oversampling filters, usually a few hundred microseconds apiece. The driver adds a safety offset so the block is never late. Then anything with lookahead in the monitoring chain contributes as well. A 128-sample buffer at 48 kHz is 2.6667 ms on paper and commonly measures 7 to 9 ms in a loopback test.

Does microsecond precision matter to a guitarist monitoring live?

Not for the feel of the performance. A player reacts to the total round trip, and a few hundred microseconds either way is inaudible — 10 ms is about the same delay as standing 3.4 m from an amplifier, which no guitarist finds strange. Microseconds start mattering the moment two signals are combined: a few sample periods of offset between a DI and a mic, or between two mics on the same cabinet, produces comb filtering you can hear plainly even though the delay itself is far too short to perceive as an echo.

Where does the latency number ASIO or Core Audio reports come from?

From the driver, in samples, which the host then divides by the sample rate to show milliseconds. An ASIO driver returns separate input and output latency values; Core Audio exposes the device latency, the stream latency, the safety offset and the buffer frame size, and the host adds them up. Both are the driver's own account of itself, so a manufacturer that omits converter delay produces a flattering figure. Trust a loopback measurement over the panel when the difference matters.