How theoretical audio latency is calculated #
One buffer stage lasts:
buffer duration in milliseconds = buffer samples ÷ sample rate × 1,000
The calculator keeps the path components separate, then adds them:
theoretical total = input buffers + output buffers + extra buffers + processing delay + known fixed delay
Input, output and extra stages are whole counts of the selected buffer size. Plug-in or processing delay is entered in samples and converted at the selected rate. Known converter, driver or device delay is entered directly in milliseconds. The total sample equivalent converts the entire sum—including fixed milliseconds—back at the selected rate, so it may be fractional.
What the path stages mean #
- Input buffer stages represent sample blocks collected before processing. Use zero for an output-only playback estimate.
- Output buffer stages represent blocks queued for playback.
- Extra or safety stages represent additional full buffers only when their existence is known or intentionally assumed.
- Processing delay represents reported plug-in delay, lookahead or another known sample-based stage.
- Known fixed delay holds delay reported in milliseconds, such as converter or device DSP time.
The output-only and input-plus-output buttons change only the three buffer-stage counts. They do not invent converter, driver, safety-buffer or plug-in values.
Buffer duration is not total latency #
A 128-sample buffer lasts approximately 2.667 ms at 48 kHz. One input plus one output buffer therefore contributes approximately 5.333 ms before any other stage. That arithmetic is exact for the stated assumptions, but the assumptions rarely describe the entire physical path.
Do not automatically double a buffer value without identifying why two stages belong in the path. Conversely, do not report a single buffer as round-trip latency when input and output buffering are both involved.
Theoretical estimates versus measured results #
This calculator does not measure an interface, driver or computer. Actual latency can also contain:
- analogue-to-digital and digital-to-analogue converter delay;
- hidden driver or hardware safety buffers;
- USB, Thunderbolt or network transport scheduling;
- operating-system and audio-engine scheduling;
- plug-in delay compensation and device lookahead;
- mixer, DSP and monitoring-path processing.
Prefer a device or audio engine's reported input and output latencies when they are trustworthy. For end-to-end verification, a loopback test captures the real path more directly. Enter a known difference as fixed delay only when its source and units are understood.
Reading changes in buffer size and sample rate #
At a fixed sample rate, doubling the buffer size doubles each buffer stage's duration. At a fixed buffer size in samples, doubling the sample rate halves its duration. Higher sample rates may increase processor load, however, and that tradeoff is not represented by the timing formula.
Changing the displayed buffer or sample rate does not change the hardware. It recalculates the hypothetical path described by the inputs.
Compare buffer sizes #
These are one buffer stage, rounded to three decimal places. They exclude converter, driver, plug-in and other fixed delays. At the same sample rate, two identical buffer stages take twice the time shown.
| Buffer (samples) | 44.1 kHz | 48 kHz | 96 kHz |
|---|---|---|---|
| 64 | 1.451 | 1.333 | 0.667 |
| 128 | 2.902 | 2.667 | 1.333 |
| 256 | 5.805 | 5.333 | 2.667 |
| 512 | 11.610 | 10.667 | 5.333 |
| 1024 | 23.220 | 21.333 | 10.667 |
At 48 kHz and 128 samples, one input plus one output buffer contributes 5.333 ms. A hypothetical additional 48 samples of processing and 1 ms of fixed delay bring that sum to 7.333 ms. Open this worked path.
The linked examples set every path input explicitly. Follow the link, then read or adjust the assumptions in the calculator. The reference table stays at the three sample rates printed above when you change the controls.
Recording versus mixing #
Playing a software instrument: an output-only example at 48 kHz and 128 samples gives 2.667 ms of output buffering. MIDI/controller timing and instrument processing can add delay; this is not the complete key-to-sound response time.
Monitoring a microphone or guitar through Live: start with the input-plus-output example, 5.333 ms of buffering. Smaller buffers reduce that contribution but can cause dropouts under load. Compare with Live's reported input/output latency and account for the actual effects in the monitoring path.
Mixing existing audio: if immediate playing response is no longer needed, a larger buffer gives the computer more time per block. At 48 kHz and 512 samples, output only, buffering contributes 10.667 ms. This changes response time, not the reverb predelay or musical delay time set inside a device. Use the BPM and delay calculator for those sound-shaping times.
In Live's Audio settings, compare the buffer choices your interface supports and use a size that plays reliably. Raising the sample rate shortens each fixed-size block, but can increase CPU demand. These examples are comparisons, not recommended settings for every computer.
With interface direct monitoring, the sound you hear bypasses Live's software-monitoring path, so the input-plus-output example does not describe that monitor mix. You also will not hear Live's effects on that direct signal. Live's Reduced Latency When Monitoring option addresses delay compensation on monitored tracks; it does not remove interface buffering or every device's processing delay.
Ableton references: How latency works, reducing latency, and monitoring options.
Common mistakes #
- Mixing samples from one sample rate with milliseconds calculated for another.
- Adding a safety buffer merely because one might exist instead of documenting the assumption.
- Treating reported plug-in delay as an extra stage when the audio engine has already compensated for it.
- Comparing a one-way playback estimate with a measured round-trip result.
- Presenting the theoretical total as a guaranteed or measured latency.