The most useful way to troubleshoot audio latency is to describe the symptom before choosing a fix. Delayed instrument monitoring, sound behind a picture, a doubled voice and crackling playback are different problems. They can coexist, but one buffer change or synchronization offset will not necessarily solve them all.
Write down what starts the event, where you hear the result, and whether the error stays constant. Then test one short, repeatable action. This approach produces better evidence than changing several settings and hoping the sound improves.
A symptom table for the first decision
| What you notice | Useful distinction | First check |
|---|---|---|
| Instrument response is late | Live monitoring delay | Input, processing and output route |
| Speech trails lips | Relative A/V offset | Which stream arrives first |
| Sync gets worse over time | Drift or changing delay | Early and late recorded events |
| Voice sounds doubled | Two audible routes or echo | Direct/software monitoring and speaker return |
| Clicks or gaps | Reliability problem | Load, buffer and device connection |
| Waveform lands late | Recording placement | Compensation and actual capture path |
The meaning of audio latency depends on the selected start and end points. Keep that definition consistent when comparing a displayed number with what you hear.
Problem 1: your own voice or instrument returns late
Check whether the signal travels through software effects before reaching your headphones. Temporarily bypass a suspect processor in a reversible project state, then repeat the same phrase. Also check the master path, not just the visible track.
If an interface supports direct monitoring, compare that route with software monitoring. Keep only one audible at a time. Direct monitoring can bypass the application’s return, but may not provide its effects and cannot substitute for generating a software instrument.
A recorded take can be aligned correctly while the performer still hears a delayed return. The music-production latency guide treats listening and recorded placement as separate questions.
Problem 2: the sound and picture disagree
Determine the direction first. If you hear a clap before seeing the hands meet, audio is early relative to video. If the visible clap comes first, audio is late. The correction must act on the appropriate stream.
Use the audio sync test for a subjective comparison, then confirm the result in the actual viewing or recording setup. Its visual-offset sign should not be copied blindly into an audio-delay control.
For example, if audio arrives at 80 ms and video at 150 ms after a source event, audio leads by 70 ms. Delaying that audio by 70 ms would align the illustrative pair. Both would still arrive after the event; synchronization is not the same as removing delay.
Problem 3: synchronization drifts during a recording
A constant error and a growing error require different investigations. Make identifiable events near the beginning and end. If an offset changes from 20 ms to 120 ms, a single 20 ms correction cannot keep the entire recording aligned.
Check device clocking, timestamps, connection stability and the application’s configuration. Nominally matching sample rates can be helpful for a coherent setup, but a mismatch does not inevitably cause drift when resampling works correctly. Do not diagnose the cause from a menu value alone.
The OBS delay and drift guide explains why short and long output checks belong in the same workflow. A preview or local headphone route may differ from the recorded program.
Problem 4: sound is doubled, hollow or echoing
Mute one monitoring route at a time. An interface’s direct sound plus the application’s delayed copy can produce doubling. In a call, a speaker feeding another participant’s microphone can return your voice after a longer journey.
These situations do not necessarily require a new driver or a negative offset. First identify the second route. Headphones can help prevent speaker-to-microphone return in a call, while correct monitoring selection can remove a duplicate local copy.
If the issue begins only when a headset microphone opens, note that exact trigger. Playback and communication modes can use different paths. The wired and wireless headphone comparison helps organize a controlled alternative-route test.
Problem 5: a smaller buffer introduces clicks
A smaller block gives the system less time to provide the next audio data. When deadlines are missed, the result may be a glitch rather than a smooth but delayed sound. Return to a stable setting and test the real workload.
One 256-sample block at 48 kHz lasts about 5.33 ms. Reducing it to 64 samples shortens that block to about 1.33 ms, but does not guarantee that the computer can sustain the new timing. The buffer-duration calculation describes one contribution, not all processing.
Check whether the problem follows a specific project, plug-in or connection. A stable empty project and an unstable dense mix are evidence of different operating demands, not proof that every setting is faulty.
Problem 6: a browser test gives no valid return
Confirm the selected microphone, permission, output route and acoustic path. Isolated headphones may not deliver the probe to the microphone. A failed detection is unavailable measurement data, not a zero-millisecond result.
The homepage sound latency test measures an acoustic return that includes output, air and input. Its separately reported browser estimates are not a physical device benchmark. Keep the microphone fixed when comparing routes.
Keep a small troubleshooting record
Record the symptom, setup and one change, then repeat the same test. Include whether the improvement survived a restart or reconnection. If a change creates another problem, restore the baseline before testing the next hypothesis.
The latency reduction checklist is useful after identifying the affected path. Stop when the setup is reliable and appropriate to your task; chasing the smallest possible number can make a usable system less dependable.
Frequently asked questions
Are crackles the same as high latency?
No. Crackles indicate an audio-quality or delivery problem. Lowering buffering can reduce waiting while making missed deadlines more audible, so evaluate stability separately.
Why is video fine but gaming feels late?
A video application may align streams by waiting. An interactive sound must follow an unpredictable action, so the same route can feel different even without a fault.
Why does an offset help briefly and then fail?
The error may be growing or changing rather than constant. Compare events at different times before applying a single fixed correction.
Can a browser tell me which component is faulty?
A complete acoustic return does not uniquely identify each component. Controlled comparisons can narrow possibilities, but routing and independent application tests remain necessary.
When should I seek device support?
If a reproducible problem persists with a supported minimal configuration, provide the manufacturer with device, driver, application and connection details plus exact reproduction steps. Avoid presenting an unrelated browser total as an isolated hardware measurement.
