Microphone and Game Audio Out of Sync in a Recording? Source Clock, Sample Rate and Timestamp Diagnosis Guide

Key takeaway
When a microphone and game audio drift apart inside the same recording, the two sources are running on different clocks or settings, so you must measure each source's offset at the start and the end of the file before changing anything.
The first sentence matters because it contains the two distinctions that decide the whole diagnosis. The first is that the two audio sources can be wrong by different amounts: the microphone might be ahead of the picture while the game audio is exactly right, or each source might be off by a different number. The second is that the error can take two time shapes — a fixed offset that is the same from the first second to the last, or a gradual drift that starts correct and grows worse as the recording goes on. Those two shapes are measured differently and repaired differently, and guessing which one applies is the most common way to waste an afternoon changing the wrong setting.
What counts as this problem — and what does not
This guide covers one narrow shape: inside a single recording file, the microphone sound and the game or desktop sound do not line up with each other. The offset is per source. One source may be early and the other on time; both may be off, but by different amounts. The mismatch is burned into the file: any player playing the same file at the same timecode hears the same mismatch. That last point is the boundary with a different class of problem in which a file plays normally in a player and only stutters inside an editor — that is a variable-frame-rate problem, not this one.
Several nearby problems look similar but are not this one:
| Symptom | What it points to instead |
|---|---|
| The whole file has one fixed offset in the same direction | A whole-file offset problem, not a per-source mismatch |
| The whole file's pitch or speed is wrong | A sample-rate labelling problem affecting the entire file |
| The recording file has no sound at all | A recording-stage problem with no audio captured |
| The recording goes silent part-way through and stays silent | A problem during recording, not a sync problem |
| Only the live stream is out of sync while the local recording is fine | A network or encoding queue problem in the stream |
| The monitoring headphones sound out of sync but the recording is fine | A monitoring-path problem, not a file problem |
| The audio and video suddenly go out of sync in a step-like jump | No external guide consulted confirms a sudden staircase shape — only a fixed offset and gradual drift; a sudden jump is most likely gradual drift noticed late, dropped frames, or a device being re-enumerated, not this guide's subject |
If the file in front of you does not match the shape in the first paragraph — two sources inside one file, wrong by different amounts — stop and re-describe it. In particular, do not treat "the audio is out of sync" as a complete description. The description is only complete when it says which source is early or late, by how much, and whether that amount is the same at the start of the file and at the end.
The clap test: how an offset is actually measured
An offset cannot be judged by ear with the precision this diagnosis needs. The practical method is a clap test. Record a short test segment in which you clap once, visibly, on camera. Then open the recording in an editor, find the frame in which the palms meet, and compare it with the sharp peak in each audio track's waveform. The difference between the visual frame and the audio peak, measured separately for the microphone track and for the game or desktop track, is that source's offset.
At 60 frames per second, one frame lasts about 17 milliseconds, because one second divided by 60 frames is roughly 17 milliseconds per frame (a figure from evezone). Counting frames therefore gives a usable estimate: if the microphone's waveform peak sits twelve frames after the palms meet, the microphone is roughly twelve frames late — about 200 ms at 60 fps. The number is an estimate, not a laboratory measurement, and it should be written down as an estimate with the frame count beside it. The essential discipline is to measure each track separately. Measuring a mixed-down file and reporting a single number hides exactly the information this guide exists to find — which source is wrong, and by how much.
That measurement can only be made if the sources are still separate. This is a hard limit, and it is stated here before any step that depends on it: if the microphone and the desktop audio were mixed down onto a single track in an MP4 file, they cannot be separated afterwards. An MP4 by default stores only the first track, so once the two sources share one track, there is no later step that pulls them apart to measure them individually. Diagnosis in that situation needs either a multi-track recording — for example, an MKV file recorded with separate tracks — or a fresh clap test recorded with the sources separated. This guide does not promise a way to un-mix a mixed-down file, because no such way exists within the tools this diagnosis uses.
The five suspect layers
There are five layers where a per-source mismatch can originate. External guides consulted for this guide emphasise different layers — one stresses the hardware clock, another stresses sample rate, a third lists three categories side by side — and they are describing different parts of the same chain, not contradicting each other. This guide therefore offers no ranking of "most common cause". What each layer explains, and the time shape it produces, is the useful distinction.
1. Source clock layer
The recording software and the microphone do not necessarily share a clock. OBS, for example, counts audio samples using its own software clock, while a microphone or a USB headset uses the hardware crystal clock on the device itself. The two clocks can disagree slightly about how long one second is. A small disagreement is invisible in the first minute and accumulates over time, which is why drift caused by this layer grows worse the longer the recording runs. A guide from freestreamhosting.org describes this mechanism as the explanation for drift that worsens with recording length rather than appearing as a fixed error from the first second.
The signature of this layer is gradual drift: correct at the start, increasingly wrong towards the end. If the opening measurement and the closing measurement differ, and the difference grows steadily, the clock layer is a strong suspect. If the offset is identical at the start and at the end, the clock layer is a weak suspect, and the fixed-delay layer below is the better place to look.
2. Sample-rate layer
Sample rate is the number of audio samples recorded per second. Two common values are 44.1 kHz and 48 kHz. If the microphone is set to 44.1 kHz in Windows while OBS is using 48 kHz, OBS must resample every frame of audio in real time, and the rounding in that resampling can accumulate into drift. A guide from Hollyland describes this drift as becoming noticeable after 20–30 minutes of recording; guides from freestreamhosting.org and cubix.design describe the same sample-rate mismatch in consistent terms. All three guides agree on the direction of the remedy: set the Windows device and OBS to the same sample rate. All three suggest 48 kHz as the shared value. That suggestion belongs to those guides; it is not a standard set by this site, and this guide presents it only as what those three guides recommend.
The signature of this layer is also gradual drift, which is why it is easily confused with the clock layer — the two are adjacent links in the same chain, and a single test cannot always separate them. The distinguishing test is the sample-rate comparison in step four: align every device and OBS to the same rate, record a fresh test segment, and see whether the drift disappears or slows substantially. If it does, the sample-rate layer was involved. If the drift is unchanged, the sample-rate layer was not the cause, and the investigation continues.
It is worth stating plainly what this layer is not. A sample-rate mismatch between two sources produces two sources drifting apart from each other, with the pitch of each source potentially sounding completely normal. A whole file whose pitch or speed is wrong throughout is a different problem — a sample-rate labelling problem affecting the entire file — and it is not this guide's subject.
3. Device timestamp layer
OBS has a setting called "device timestamp", which tells OBS to use the hardware device's clock instead of its own software clock for that device. Two external guides, from freestreamhosting.org and cubix.design, both state that turning this setting off gives the correct result, on the grounds that the hardware clock and OBS's internal clock run at different speeds — a difference of even 0.01% can accumulate to hundreds of milliseconds over an hour, according to freestreamhosting.org and cubix.design. However, some users report that turning it on gives the correct result for them. The external guidance is inconsistent, and this guide therefore treats the setting as a comparison test, not as a question with a fixed correct answer.
The test is straightforward: turn the setting off, record a fresh test segment, and compare the drift with the previous recording. If the drift improves, the timestamp layer was involved. If it does not improve, that layer was not the cause. What must not be done is to write down one position — on or off — as correct for every setup. The honest record is the comparison result: which position was used, and what the drift measured under it.
One external guide contains its author's own percentage claim about how many users this setting affects. That figure is the author's claim, it has not been verified by this site, and this guide does not repeat it. No percentage in this guide should be read as this site's finding.
4. Fixed-delay layer
Some sources add a fixed processing delay. An HDMI capture card, or an external camera source such as an AirPlay camera, takes time to process the picture before it reaches the recording software. The microphone, going directly into OBS, has no such delay, so the microphone ends up ahead of the picture by a fixed amount. One external guide, freestreamhosting.org, describes capture-card delay as roughly 50–500 milliseconds; that range is that guide's description, not a standard set by this site. Bluetooth devices add their own delay: a guide from Hollyland describes Bluetooth delay as roughly 100–300 milliseconds, again as that guide's description, not this site's standard. A Bluetooth hands-free profile can also change the sample rate and the delay, which is why the hands-free setting deserves its own check — many users do not realise they are using the hands-free profile at all.
The signature of this layer is a fixed offset: wrong by the same amount from the first second, not growing worse over time. If the opening measurement and the closing measurement are the same number, the fixed-delay layer is a strong suspect. The remedy is not a single global adjustment, because each source has its own offset. A guide from Hollyland states this directly: each source has its own independent offset field, and only the source that is wrong should be adjusted, not all of them at once.
5. Monitoring and buffer layer
Windows exclusive mode allows one program to control an audio device directly, bypassing the Windows audio mixer. When OBS and another program compete for the same device, the audio buffer can back up, and the audio timing can shift unpredictably. A guide from Hollyland describes this behaviour, and freestreamhosting.org describes it consistently. The same guide notes that when the buffer reaches its limit, OBS can show a "max audio buffering reached" warning, and the audio begins to drift.
There is a second, subtler point in this layer, and it is a trap. The audio heard through monitoring headphones and the audio written into the recording file travel along different timing paths, and they can drift independently of each other. A guide from freestreamhosting.org states this directly: the recording can be correct while the monitoring drifts. That means hearing a mismatch in the headphones does not prove that the file is mismatched. Whether the file has a problem can only be judged from the recording file itself, never from the monitoring. Anyone who treats "I heard it out of sync" as proof that the file is out of sync has confused the monitoring path with the recording path, and will chase a file problem that does not exist.
The five decisive forks
Five comparisons do the diagnostic work. Each one is a fork: the result sends the investigation down one branch and closes off the other.
First fork — per-track offset measurement. Using a multi-track recording or a clap test, measure each track's offset against the picture separately. The microphone off by some amount while the desktop audio is on time means the microphone source is the problem. Both tracks off by the same amount means a whole-file offset, which is a different problem from this guide's subject.
Second fork — fixed versus gradual. Measure once at the start of the file and once at the end. The same difference at both ends means a fixed offset, pointing to the fixed-delay layer. A difference that grows towards the end means gradual drift, pointing to the clock or sample-rate layers. A guide from freestreamhosting.org states that distinguishing these two shapes is the first step of diagnosis, because confusing them leads to the wrong remedy.
Third fork — per-device sample-rate comparison. Set the Windows recording device and OBS to the same sample rate, then record a fresh test segment. Drift that disappears or slows substantially means the sample-rate layer was involved. Drift that is unchanged means it was not.
Fourth fork — device timestamp on/off comparison. Turn the device timestamp setting off and record again. Drift that improves means the timestamp layer was involved. No improvement means it was not. There is no standard answer to fall back on; only the comparison result counts.
Fifth fork — isolating sources one by one. Record with only the microphone, then only the desktop audio, then with a wired microphone instead of a Bluetooth one, then with the Bluetooth hands-free profile disabled. Whichever change makes the sync correct identifies the offending source. The hands-free profile deserves explicit mention, because many users do not realise they are using the hands-free profile.
Seven-step diagnosis
Follow the steps in order. Each step depends on the one before it, and changing several settings at once destroys the comparison that would have identified the layer.
Step 1 — Define the shape. Confirm that two audio sources inside the same file disagree with each other. Write down which source is early or late, and by how much, using a clap test rather than an impression. "Out of sync" is not a measurement; a per-source offset in milliseconds, with the frame count beside it, is.
Step 2 — Clap test. Record a test segment with a visible clap. In the editor, find the frame where the palms meet and compare it with the sharp peak in each audio track's waveform. Measure each track's offset against the picture separately. At 60 frames per second, one frame is about 17 milliseconds, so the frame count converts directly into an approximate offset.
Step 3 — Start versus end comparison. Measure the offsets once near the start of the file and once near the end. The same difference at both ends means a fixed offset; a difference that grows means gradual drift. This single comparison decides whether the fixed-delay layer or the clock and sample-rate layers are the better suspects, and it must be made before any setting is changed.
Step 4 — Per-device sample-rate comparison. Set the Windows recording device and OBS to the same sample rate — the three external guides consulted all suggest 48 kHz as the shared value — and record a fresh test segment. If the drift disappears or slows substantially, the sample-rate layer was involved. If the drift is unchanged, it was not, and the investigation continues.
Step 5 — Device timestamp on/off comparison. Turn the device timestamp setting off and record again. Compare the drift with the previous recording. Improvement means the timestamp layer was involved; no improvement means it was not. Record which position was used and what the drift measured under it, because there is no standard answer to rely on instead.
Step 6 — Isolate sources one by one. Record with only the microphone, then only the desktop audio, then with a wired microphone instead of a Bluetooth one, then with the Bluetooth hands-free profile disabled. Whichever change makes the sync correct identifies the offending source. Change one thing at a time; changing two sources together makes the result unreadable.
Step 7 — Sync Offset comparison, then stop. OBS provides a Sync Offset field for each source, and adjusting it is a comparison test, not a repair. Finding a number that makes the test segment line up does not mean the cause has been found; it means the symptom has been compensated for in that segment. If the cause still cannot be separated after these steps, stop changing settings. Bring the records: each track's offset, the sample-rate settings, and the timestamp setting's on/off state. The shop is in Shek Mun, Dali Plaza, Room 1209; the phone number is 2186 8010 and WhatsApp is 6558 6806. The published inspection charges are HK$400–2,800 across three tiers.
Six common misconceptions
1. "Two sources out of sync means the whole file is offset." Wrong. A per-source offset and a whole-file offset are two different shapes with different measurements. Both tracks off by the same amount is a whole-file problem; one track off while the other is correct is a per-source problem, and only the per-source measurement can tell them apart.
2. "A sample-rate mismatch means the whole file changes pitch." Wrong. A whole-file pitch or speed error is a different problem. In this guide's problem, two sources drift apart from each other while the pitch of each source may sound completely normal.
3. "Sync Offset is the fix." Wrong. Sync Offset is a comparison test. Adjusting the number until a test segment lines up does not identify the cause, and it must not be presented as having solved the problem.
4. "The device timestamp setting has one correct answer." Wrong. External guidance is inconsistent, with some users reporting that turning it on works for them. The setting must be treated as a comparison — off versus on, with the drift measured under each — not as a rule.
5. "Hearing it out of sync in the headphones means the file is out of sync." Wrong. Monitoring and recording travel along different timing paths and can drift independently. Whether the file has a problem can only be judged from the recording file itself.
6. "A mixed-down MP4 can be separated afterwards to measure each source." Wrong. An MP4 by default stores only the first track, and once the microphone and desktop audio share one track, they cannot be pulled apart. Measuring per-source offsets requires a multi-track recording or a fresh clap test with the sources separated.
What to preserve before seeking help
Do not change the sample rate, the timestamp setting, and the Sync Offset all at once in the hope that one of them will fix the problem. A combined change that happens to improve the sync proves nothing about which setting mattered, and the next recording may drift again for an unidentified reason. Keep the original recording, keep a written list of each track's offset at the start and at the end of the file, the sample-rate settings on each device, and the timestamp setting's on/off state, and keep the test recordings that show the comparison results. That list is what allows the next person to continue the diagnosis from the measurements instead of starting again from an impression.
FAQ
Why is my microphone out of sync with the game audio, but the game audio matches the picture?
Because the two sources run on different clocks or settings. The microphone uses its own hardware clock, while the recording software uses its own software clock, and a sample-rate mismatch between them can add drift. Each source must be measured separately against the picture; a single "out of sync" figure cannot describe two sources that are wrong by different amounts.
How do I tell whether the problem is a fixed delay or gradual drift?
Measure the offset near the start of the file and again near the end. The same difference at both ends means a fixed offset, which points to the fixed-delay layer such as a capture card or Bluetooth. A difference that grows towards the end means gradual drift, which points to the clock or sample-rate layers.
Should I set everything to 48 kHz?
Three external guides consulted for this guide all suggest aligning the Windows device and OBS to the same sample rate, and all three suggest 48 kHz as the shared value. That suggestion belongs to those guides, not to this site. The honest way to use it is as a comparison test: align the rates, record a fresh test segment, and see whether the drift disappears or slows.
Should the device timestamp setting be on or off?
There is no standard answer. Two external guides state that turning it off gives the correct result, but some users report that turning it on works for them. Treat it as a comparison: turn it off, record again, and compare the drift. Record which position was used and what the drift measured under it.
Can I fix the sync in a recording where the microphone and game audio are already mixed into one MP4 track?
No, not by separating them afterwards. An MP4 by default stores only the first track, and once the two sources share one track they cannot be pulled apart to measure each source individually. Diagnosis needs a multi-track recording or a fresh clap test with the sources separated; the mixed-down file itself cannot be un-mixed.
I adjusted Sync Offset and the test clip lines up. Is the problem fixed?
Not necessarily. Sync Offset compensates for the symptom in that segment; it does not identify the cause. The cause is only found when the clock, sample-rate, timestamp, or fixed-delay layer that produced the offset has been identified by comparison. If the underlying layer is still wrong, the drift can return in a longer recording.