Audio latency is the time delay between a signal entering an audio system and the same signal leaving it, measured in milliseconds or in samples. In digital audio, latency accumulates at every stage that stores, converts or processes samples: the analog-to-digital converter, the driver and audio buffer, the host software, any plugin with look-ahead, and finally the digital-to-analog converter feeding speakers or headphones.
Latency matters most when a performer hears their own sound in real time. A guitarist playing through a software amplifier or a singer monitoring with reverb notices delay long before a listener of finished playback would. For that reason engineers distinguish round-trip latency, which a performer experiences, from output latency, which only affects how quickly playback reacts to the transport controls.
Where audio latency comes from
Audio latency is the sum of several independent delays, and each one has a different cause. The largest contributor in a typical computer setup is the buffer size, because the processor receives audio in blocks rather than one sample at a time. A buffer of 256 samples at 48 kHz holds about 5.3 ms of audio, and the system needs at least one input buffer and one output buffer before any sound returns.
Converters add a smaller but fixed delay. Modern sigma-delta converters use oversampling and digital decimation filters, which typically cost between 0.3 ms and 1 ms per direction depending on the filter design and sample rate. Drivers and the operating system may add safety buffers that are not visible in the host settings, which is why reported and measured values often differ.
The table below lists the stages of a typical monitoring path and the order of magnitude each one contributes.
| Stage | Typical delay | What controls it |
|---|---|---|
| Analog-to-digital conversion | 0.3 to 1 ms | Converter filter design and sample rate |
| Input buffer | 0.7 to 21 ms | Buffer size divided by sample rate |
| Plugin processing | 0 to over 100 ms | Look-ahead and linear-phase algorithms |
| Output buffer | 0.7 to 21 ms | Buffer size divided by sample rate |
| Driver safety offset | 0 to several ms | Driver model and manufacturer |
| Digital-to-analog conversion | 0.3 to 1 ms | Converter filter design and sample rate |
Calculating latency from buffer size and sample rate
The basic formula is simple: latency in seconds equals samples divided by sample rate. A buffer of 128 samples at 44.1 kHz therefore lasts about 2.9 ms, while the same 128 samples at 96 kHz last about 1.3 ms. Doubling the sample rate halves the time per buffer, but it also doubles the processing load, so the practical gain is often smaller than the arithmetic suggests.
The following table shows one-way buffer durations for common settings. Round-trip latency is usually at least twice the buffer value plus converter and driver delays.
| Buffer size | At 44.1 kHz | At 48 kHz | At 96 kHz |
|---|---|---|---|
| 32 samples | 0.73 ms | 0.67 ms | 0.33 ms |
| 64 samples | 1.45 ms | 1.33 ms | 0.67 ms |
| 128 samples | 2.90 ms | 2.67 ms | 1.33 ms |
| 256 samples | 5.80 ms | 5.33 ms | 2.67 ms |
| 512 samples | 11.61 ms | 10.67 ms | 5.33 ms |
| 1024 samples | 23.22 ms | 21.33 ms | 10.67 ms |
Sound in air travels roughly 34 cm per millisecond, which gives a useful sense of scale. A round trip of 6 ms is comparable to standing about two meters from a guitar amplifier, a delay most players accept without noticing.
How much latency performers tolerate
Tolerance depends on the instrument and on what the performer hears. Vocalists are the most sensitive, because they hear their own voice through bone conduction with no delay at all; a delayed headphone signal combines with it and produces comb filtering or a phasing sensation at only a few milliseconds. Drummers and percussionists notice timing offsets in transient-heavy material, while pad and string players are usually more forgiving.
| Use case | Comfortable round trip | Main concern |
|---|---|---|
| Vocal monitoring | Below about 3 to 5 ms | Phasing against bone conduction |
| Drums and percussion | Below about 6 to 8 ms | Timing feel on sharp transients |
| Guitar through amp simulation | Below about 8 to 10 ms | Responsiveness of picking |
| Software synthesizer playing | Below about 10 ms | Connection between key and sound |
| Mixing and playback | Tens of milliseconds acceptable | Transport reaction only |
These ranges are practical guidelines rather than fixed thresholds. Individual perception varies, and an experienced player may adapt to a constant delay that would disturb a beginner. What performers find hardest is inconsistent latency, where the delay changes between sessions or drifts during a take.
Audio latency compared with related concepts
Several neighboring terms in digital audio describe parts of the same problem. The audio buffer is the mechanism that creates most latency, while ASIO and similar low-latency driver models are the tools that keep buffers small and the path short. Plugin delay compensation is a different matter: it aligns tracks during playback but cannot remove delay from live monitoring.
| Concept | Relation to latency | Affects live monitoring |
|---|---|---|
| Audio buffer | Main source of delay | Yes |
| ASIO and Core Audio | Driver paths that minimize delay | Yes |
| Plugin delay compensation | Aligns tracks after processing | No, only playback alignment |
| Jitter | Timing variation of the sample clock | No, affects conversion accuracy |
| Direct monitoring | Bypasses the computer entirely | Removes software delay |
Reducing latency in practice
The most effective step is to lower the buffer size while recording and raise it again for mixing, when processing power matters more than responsiveness. Engineers also use direct monitoring on the audio interface for vocals, so the performer hears the input before it reaches the computer, and add effects only to the recorded track afterward.
- Use a dedicated low-latency driver instead of generic system audio drivers on Windows.
- Disable look-ahead plugins such as linear-phase equalizers and mastering limiters during tracking.
- Freeze or render heavy tracks to free processor headroom for a smaller buffer.
- Measure real round-trip latency with a loopback cable instead of trusting the displayed value.
- Set recording offset compensation so recorded takes line up with existing material.
The main trade-off is stability: a buffer that is too small causes clicks, dropouts and crackles when the processor cannot finish a block in time. The right setting is the smallest buffer that runs without dropouts under the actual session load.
In plain words
Audio latency is the short wait between making a sound and hearing it come back out of a computer. The computer collects sound in small batches before it can work on it, and each batch takes a moment to fill. Smaller batches mean less waiting but more work for the computer.
Think of a bucket brigade passing water. Small buckets move water quickly but need many fast hands; large buckets are easier but slower to arrive. The goal is the smallest delay the computer can handle reliably, so musicians feel connected to their own sound.
A real-world example
A home studio owner records a singer through a USB interface at 48 kHz with a buffer of 1024 samples, left over from a large mixing session. The singer complains that her voice in the headphones sounds hollow and late. A loopback measurement shows a round trip of about 46 ms, far too long for vocal monitoring.
The engineer drops the buffer to 64 samples, disables a linear-phase equalizer on the master bus, and switches the interface to direct monitoring with a light reverb added on a separate return. The measured round trip falls to roughly 4 ms, the singer stops hearing the doubled effect, and the takes become noticeably more confident. For the mix the next day, the buffer goes back to 1024 samples to give the plugins room to work.