What Audio Latency Really Is, and How to Fix It

Latency is the gap between making a sound and hearing it back. Small delays feel instant; larger ones make playing along impossible. Here is what creates the delay and how to keep it low.

Published ; last updated
LatencyRecordingTroubleshooting

Latency is the single most misunderstood source of frustration in home recording. You sing or play, you hear the result a moment later, and everything feels disconnected. Sometimes the delay is so small you cannot consciously notice it; sometimes it is large enough to make playing along impossible. Understanding where the delay comes from makes it far easier to fix.

The signal path from microphone to ear

When you make a sound, the microphone converts it to an electrical signal. The audio interface or built-in sound card converts that signal to digital numbers. Those numbers sit in a buffer until the computer asks for them. The computer processes them through the browser or DAW, mixes them with anything already playing, puts the result into another buffer, and sends it back out to the headphones. Each of those steps takes a small amount of time. Added together, that is latency.

The total delay you experience while playing is called round-trip latency. It includes the input path, the processing time, and the output path. If the software does not compensate correctly, a newly recorded track will also sound late relative to the existing tracks, because the input was captured after the performance actually happened.

Buffer size and sample rate

The biggest lever you can pull is the buffer size. The buffer is a small bucket of samples that the audio system fills before it hands the audio over. A smaller buffer means the system processes audio in smaller chunks, so the delay is shorter. A larger buffer gives the computer more time to finish each chunk, which reduces the risk of dropouts and clicks when the CPU is busy.

At a 44.1 kHz sample rate, a buffer of 128 samples lasts about 2.9 milliseconds. A buffer of 512 samples lasts about 11.6 milliseconds. A buffer of 1024 samples lasts about 23 milliseconds. Those numbers are only the output side. For round-trip latency you can roughly double them, because the input buffer adds a similar delay. At 1024 samples, round-trip latency is likely somewhere in the 30 to 50 millisecond range once you include the rest of the audio path.

Sample rate also matters because it changes how long each sample lasts. At 96 kHz, a 128 sample buffer is only about 1.3 milliseconds. The catch is that higher sample rates use more CPU, so you may need a larger buffer to avoid dropouts. The net improvement is not always as large as the numbers suggest.

Input, output, and round-trip latency

Input latency is the time between the sound entering the microphone and it becoming available to the software. Output latency is the time between the software producing a sample and that sample reaching your ears. Round-trip latency is the sum of the two plus any processing delay in between.

Different audio APIs report different numbers. On macOS, Core Audio usually gives very low latency with stable performance. On Windows, ASIO drivers are the standard for low latency, while the default Windows audio path can be surprisingly sluggish. On Linux, PulseAudio or PipeWire may add extra buffering unless the browser is using a low-latency JACK or ALSA path. In the browser, the Web Audio API exposes a latency hint and a reported base latency, but the actual figure depends on the operating system and the hardware.

Why browsers differ from native DAWs

A native DAW running with a dedicated ASIO or Core Audio driver can often reach total round-trip latencies below 10 milliseconds. A browser has an extra layer of sandboxing and scheduling. The browser asks the operating system for audio buffers, but it does not control the OS audio stack directly. That means browser-based recorders are rarely as low-latency as a well-tuned native DAW.

The good news is that you do not need zero latency to record a good take. You only need the delay to be small enough that your brain does not fight it. Under about 10 milliseconds, most people can play naturally. Between 10 and 20 milliseconds, some players feel a slight detachment. Above 20 to 30 milliseconds, especially on percussive instruments such as drums or picked acoustic guitar, the delay becomes genuinely distracting.

Latency compensation

Latency compensation is the trick that makes overdubs line up even when round-trip latency is not zero. The software measures or estimates the total delay, then trims the same number of samples from the start of the recorded take. The take still sounds late while you are playing, because you are hearing the mix through the output buffer. But once the recording is finished, the file is shifted earlier so it lines up with the existing tracks.

Compensation is not perfect. If the reported latency is wrong, the take ends up slightly early or late. The reported latency may also vary between devices and drivers. That is why NovaStudio provides a manual latency offset slider. If your overdubs consistently arrive late or early, you can nudge the timing by a few milliseconds until everything locks together.

Measuring your own round-trip delay

You can measure latency with a simple loopback test. Send the output of your interface into its own input using a cable. Record a short click or impulse, then measure the gap between the original click and the recorded click. The difference is your round-trip latency. If you do not have a loopback cable, you can also record a hand clap while monitoring through headphones. This is less precise, but it is good enough to spot a problem in the tens of milliseconds.

Once you know the round-trip latency, you have a useful reference, but you rarely need to type it into the manual offset. NovaStudio already measures and compensates for the estimated round-trip delay, so the slider is mainly for trimming the last few milliseconds of residual drift. Record a test take, listen to how it lines up with the guide track, and adjust in small steps. 5 milliseconds is about 1.7 metres of sound travel, so tiny changes are audible on fast transients.

Practical thresholds

  • Under 10 ms: feels immediate; comfortable for almost any instrument.
  • 10-20 ms: acceptable for most playing; may feel slightly soft on drums and fast picking.
  • 20-30 ms: noticeable; usable for vocals and slow parts, harder for tight timing.
  • Above 30 ms: difficult to play along with; worth investigating buffer size, drivers, or a different interface.

Keeping latency low in practice

Use the smallest buffer size that does not produce clicks. Close background applications that use the CPU or audio device. Use headphones rather than speakers, both to avoid feedback and because speakers add extra acoustic delay. If you are using a built-in sound chip, consider a dedicated USB audio interface with its own ASIO driver on Windows, or a class-compliant interface on macOS and Linux. Record with monitoring turned off in the software if you can monitor directly from the interface instead, because that removes the output side of the round trip entirely.

Latency is not something to fear. It is something to measure, understand, and compensate. Once you do, recording in a browser or a DAW becomes much the same problem: get the buffer size right, check the offset, and trust your ears.