Recording File Audio Stops Halfway, Sound Then Silent? Encoder, Audio Device and Track Diagnosis Guide

2026-11-18 | GPU Repair | 44 minutes | HK GPU Fix Editorial Team

Recording File Audio Stops Halfway, Sound Then Silent? Encoder, Audio Device and Track Diagnosis Guide

Key Summary

A recording that starts with sound and goes silent halfway through holds two kinds of silence: a track that stopped being written, and a track written with silence in it. This guide uses the exact time the sound stops, a MediaInfo track-length comparison, and a re-recorded test clip to separate an encoder fault from a device or source fault.

The Shape of the Problem: Sound Present, Then Gone

This guide is about one narrow shape, and that shape is the whole diagnosis. A recording is made — in OBS, in NVIDIA ShadowPlay, in the Xbox Game Bar, or in AMD ReLive — and on playback the picture runs normally from the first second to the last. The sound starts normally as well: the game is audible, the microphone is audible, the first minutes sound exactly as they should. Then, partway through the file, the sound stops. It does not fade, crackle, or drop in and out — it stops completely and stays stopped until the end of the file. The picture keeps going. The file plays to the end.

Three facts pin the shape down, and all three have to hold. First, sound is present at the start — a recording silent from the first second belongs with a different guide. Second, the silence starts at one identifiable point in time and then runs to the end of the file — a recording whose sound comes and goes, or crackles and breaks up, is a different shape. Third, the picture stays normal throughout — a recording whose picture breaks too belongs with a different guide. Without all three, the diagnosis has no anchor, and the wrong road gets followed.

The most important fact about this shape has to be stated before anything else, because it changes how the whole guide has to be read. No outside source consulted for this guide directly documents this exact shape — a single continuous local recording file, sound at the start, sudden silence partway through — with an established cause. The five suspicion layers in this guide are hypotheses to be tested by comparison, not conclusions. They are the most reasonable places to look, drawn from sources that describe neighbouring shapes, and each one has to be proved or excluded by a test before it can be named as the layer. A guide that wrote any one of them as a settled answer would be inventing a cause the sources do not give.

That honesty gap is also why the exact time the sound stops matters so much. The time code at which the sound disappears is itself evidence. Every layer in this guide is checked against that one number: the minute and second at which the file goes silent. If the silence begins at the same time as a scene change in the recording software, that points at the source layer. If it begins at the same time as a Windows power event or a device re-enumeration, that points at the device layer. If the audio track in MediaInfo is shorter than the video track by exactly the amount of silence at the end, that points at the encoding or writing layer. The time code is the one piece of evidence that exists before any setting is changed, and it has to be written down first, because every comparison after it is read against it.

The shape also has to be separated from its neighbours at the start, because several of them look similar on a first viewing. One neighbouring shape is a recording that is silent from the first second to the last — that is covered in the guide numbered 503, and its shape is total silence, not silence that begins partway through. Another neighbour is a recording whose sound is present all the way through but has changed pitch or speed — that is covered in the guide numbered 504, and the two guides are separated by one fact: in this guide the sound disappears, in that guide the sound is present but wrong. A third neighbour is a recording whose sound crackles and breaks up when the machine is under load in a game — that is covered in the guide numbered 311, and its shape is degraded sound that is still there, not silence. A fourth neighbour is a recording whose sound drifts further and further out of sync with the picture as the file goes on — that is covered in the guide numbered 496, and its shape is a track that exists but is in the wrong place in time, not a track that has gone silent. A fifth neighbour is a recording file that plays with sound in VLC but shows no audio track at all when it is imported into an editing program — that is covered in the guide numbered 506, and the difference is decisive: in this guide the file itself is silent in the second half, in that guide the file has sound and the editing program is not reading it. The boundary section near the end of this guide lists every one of these neighbours with the exact shape that belongs to it.

Before any test is run, write down the baseline: the recording tool and its version, the output settings, the file format, the file length, and the exact time code at which the sound stops. Without it, "it went silent halfway through" is a feeling, not evidence.

How the Recording Audio Chain Runs in Five Segments

A recording file's sound has to travel through five segments before it lands in the file, and any one of them can break in a way that produces exactly this shape: sound at the start, silence after some point, picture normal throughout. The segments are the source, the device, the capture, the encoder, and the container. "It went silent halfway through" on its own cannot say which segment broke. The segments have to be compared one by one, and the comparison has to be read against the time code at which the sound stopped.

The first segment is the source. The game sound and the system sound and the microphone are separate sources, each with its own volume and its own mute switch inside the recording software. A media source can also be set to stop when it finishes playing, or to close its file when it is idle, and a scene change can switch to a scene that does not contain the sound source at all.

The second segment is the device. Windows gives every audio device an internal identifier, and when a device is re-enumerated — after a system update, a power event, being unplugged and plugged back in, or a change behind a hub — the device can receive a new identifier. Recording software that is bound to the device by its identifier will then fail to find the device it was using, and the input goes inactive or falls back to a default that may not be the intended microphone. Separately, a USB audio device can be put into a low-power state by USB selective suspend: Windows still sees the device, but the device no longer produces sound. These are two different device-layer events — the device going to sleep, and the device disappearing under a new identifier — and they are checked in different ways, which is why the guide keeps them apart.

The third segment is the capture. Recording software binds a source either to a named device or to the "default device". When it is bound to a named device and that device disappears, the source goes inactive. When it is bound to the "default device", the recording software decides which device is the default when it starts, and it does not follow Windows if Windows changes the default afterwards. That is a detail worth stating plainly, because it is counter-intuitive: changing the default device in Windows during a recording does not make the recording software follow the change. It keeps using the device that was the default when the recording started.

The fourth segment is the encoder. The audio encoder compresses the sound into a format such as AAC and writes it into the container. If the encoder is overloaded, if the audio buffer fills up, or if the disk cannot keep up with the write, the audio track can stop being written while the video track continues. The file then has a video track that runs to the end and an audio track that stops early. That is one of the two kinds of silence this guide is built around: the track stopped being written. It is also the reason the MediaInfo comparison matters so much — an audio track shorter than the video track is the fingerprint of this layer.

The fifth segment is the container. A container such as MP4 or MKV holds the video track and the audio track together in one file, and it writes its index when the recording is finished properly. The traditional MP4 format is not crash-safe: if the recording is interrupted by a crash, a power cut, or the program being closed, the file may never be finished properly, and the audio track's index may be incomplete, so a player reaches a point in the file and has no more audio to play. MKV is more crash-safe in this respect. The container is also why the file format has to be written down in the baseline: the same symptom in an MP4 and in an MKV points at different layers.

Three Pairs of Concepts That Must Not Be Mixed

Beyond the five segments, three pairs of concepts have to be kept apart, because mixing them is how this diagnosis goes wrong.

The first pair is silence written in, against nothing written at all. An audio track whose length matches the video track, but whose waveform in the second half is flat zero, is "silence written in": the encoder kept writing, and what it wrote was silence. An audio track shorter than the video track is "stopped writing": the encoder stopped, and there is nothing after that point. The two point at different layers — the first at the source or device layer, the second at the encoding or writing layer — and they cannot be treated as the same thing.

The second pair is a device going to sleep, against a device disappearing. USB selective suspend puts a device into a low-power state: Windows still sees it, but it no longer produces sound. Re-enumeration gives the device a new internal identifier: the recording software, bound to the old identifier, can no longer find it. Both are device-layer events, but they are checked differently: the first by looking at the power settings, the second by looking at Device Manager and at how the recording software binds its sources. A guide on obscountdown.com describes the re-enumeration mechanism in detail and advises switching off USB selective suspend and binding by device name. That guide describes devices disappearing before a recording starts, not during one, so applying it to the mid-recording shape is an inference, and this guide marks it as one.

The third pair is gradual misalignment, against sudden silence. A mismatched sample rate causes gradual drift: the sound gets further and further out of sync as the file goes on, but it does not suddenly go silent. Sudden silence is a switch-like event: sound at one time code, nothing after it. This distinction is the reason this guide never writes sample-rate drift as a cause of mid-file silence. Drift is a gradual error. Sudden silence is an on-off event.

The Honesty Gap, Stated Plainly

The five suspicion layers below are hypotheses. No outside source consulted for this guide directly documents a single continuous local recording file, sound at the start, sudden silence partway through, with an established cause. The sources describe neighbouring shapes: a device that disappears before a recording starts, a track that was never assigned to the recording, a meter that moves while the file is silent, a recording that is silent from the start. Each of those is useful, and each one is marked in this guide with the shape it actually describes, so that no neighbouring shape is presented as proof of this one.

The Deciding Forks: Three Comparisons That Narrow the Field

Three comparisons carry the whole diagnosis, and they have to come before any setting is touched. Each one rules out a whole set of possibilities at once, so the order matters: change a setting first, and the evidence that comparison would have given is gone for good.

Fork One: The Time Code at Which the Sound Stops

The first fork is the time code itself. Play the file in VLC, section by section, and write down the exact minute and second at which the sound disappears. Then open the file in MediaInfo and compare the audio track length against the video track length. If the audio track is shorter than the video track, the sound stopped being written at that point, and the fault is in the encoding or writing layer. If the audio track is the same length as the video track, but the second half of its waveform is silence, the sound kept being written and what was written was silence, and the fault is in the source or device layer, or in the track layer. This fork is the one that decides which of the two kinds of silence the file holds, and it is the reason the time code has to be written down before anything else is tested.

Fork Two: Did the Volume Meter Stop, or Keep Moving?

The second fork can only be answered by watching the meter, which means it can only be answered by a re-recorded test, because the meter state during the original recording cannot be recovered afterwards. Record a test clip of the same length, with the same settings, and watch the volume meter all the way through, noting the time code at which it stops, if it stops. If the meter stops, the sound stopped reaching the recording software, and the fault is in the device or source layer. If the meter keeps moving but the file is silent in the second half, the sound reached the software and was lost at the step where it is written into the file — the encoding, writing, or track layer.

Fork Three: VLC Against a Second Player

The third fork is the player comparison. Play the same file in VLC and in a second player, and compare the two at the same time code. If both players go silent at the same time code, the silence is in the file itself, and the fault is in the file layer. If VLC still has sound in the second half while the second player is silent, the fault is in how the file is being read — and the road is the one in the guide numbered 506, not this one. VLC plays only the first track by default, so sound in the second half may be coming from a different track entirely. "VLC plays the second half" does not mean the file is fine.

ForkComparisonResult AResult B
Time code comparisonPlay the file in VLC section by section, write down the exact time the sound stops, compare MediaInfo audio track length against video track lengthAudio track shorter than the video track — encoding or writing layerAudio track the same length, second half silent — source, device, or track layer
Volume meter comparisonRe-record a test clip of the same length and watch the meter throughoutMeter stops — device or source layerMeter keeps moving but the file is silent in the second half — encoding, writing, or track layer
Player comparisonPlay the same file in VLC and in a second player, and try each audio track in VLC's track menuBoth players silent at the same time code — the silence is in the fileVLC has sound in the second half, the second player does not — reading layer, follow the guide numbered 506

The Five Suspicion Layers

The five layers below are the whole map of this symptom. They are ordered from the device end of the chain to the file end, not by how likely they are, and the order itself is part of the diagnosis. The sources consulted for this guide disagree about which layer to suspect first, and no source names a leading cause for this shape, so this guide never writes that any one layer is the most likely. All three directions have to be compared. The time code comparison is first because it costs almost nothing and it separates the two kinds of silence. The MediaInfo comparison is second because its result decides whether the sound was written at all.

LayerWhat it isDecisive test
First layer: audio deviceA USB audio device falls into a low-power state, or Windows re-enumerates a device and gives it a new internal identifier the recording software cannot findDevice Manager comparison and power settings check, read against the time code
Second layer: recording sourceA source is muted, a scene change switches the active scene, or a media source stops when it finishes playing or closes its file when idleRe-recording test watching the source meters, scene change time against the silence time code
Third layer: track assignmentA source is not assigned to the track being recorded, or is assigned to a second track while the player only plays the firstAdvanced Audio Properties track grid, MediaInfo track count, VLC track menu
Fourth layer: encoder and writingThe encoder is overloaded, the audio buffer fills up, the disk cannot keep up, and the audio track stops being written while the video track continuesMediaInfo audio track length against video track length
Fifth layer: file and containerThe recording was interrupted by a crash or a power cut, the MP4 was never finished properly, and the audio track index is incompleteMediaInfo cannot read the full audio track information; VLC against a second player; MP4 against MKV

First Layer: Audio Device

This is the layer the device comparison tests. A USB audio device can be put into a low-power state by USB selective suspend: Windows still sees the device, but the device no longer produces sound. Separately, Windows can re-enumerate a device — after a system update, a power event, being unplugged and plugged back in, or a change behind a hub — and give it a new internal identifier. Recording software bound to the device by its identifier will then fail to find the device it was using, and the input goes inactive. A Microsoft Q&A thread on learn.microsoft.com — a discussion thread of user reports, not an official Microsoft document — records several users reporting USB audio devices failing after Windows 11 24H2, possibly connected to Windows power management putting USB devices into a low-power state, with the same advice to switch off USB selective suspend.

The check for this layer is a direct one. Open Device Manager and look at whether the audio device is still there and whether it carries a yellow warning. Check the Windows power settings for USB selective suspend. Check how the recording software binds its sources — by device name, or by "default device". And write down the result, rather than relying on a feeling that "the device is fine". The decisive comparison is the time code: if the silence begins at the same time as a Windows power event or a device re-enumeration, that is the layer.

Second Layer: Recording Source

This is the layer the re-recording test watches for. Inside the recording software, each source has its own mute switch, and a scene change partway through a recording can switch the active scene to one that does not contain the sound source. A media source can be set to stop when it finishes playing, or to close its file when it is idle. Any of these can produce silence that begins at a single identifiable time code and then continues to the end of the file. A guide on umatechnology.org describes the same shape for streaming — sound stopping when the video changes — and advises checking the volume meter at four points (the recording software meter, the local recording, the headphones, and the stream), and checking in the advanced audio properties whether each source is assigned to the track used for output.

The monitoring mode deserves a mention here, because it is a silent switch that produces exactly this kind of confusion. A monitoring mode set to "Monitor Only (mute output)" deliberately makes the output silent while the meter still moves.

Third Layer: Track Assignment

This is the layer the track comparison tests. OBS can record up to six independent audio tracks at once, and each source has to be assigned separately to decide which tracks it is written into. If a source is not assigned to the track the recording output is set to record, or is assigned to a second track while the player only plays the first, the file can have sound at the start — from a track that is being played — and silence afterwards, if the sound after that point was only ever written into a track the player never plays. A guide on geekchamp.com puts this plainly: a meter that moves while the recording is silent points at a source that was never assigned to the recording track, or at a mute further down the chain.

The check is to open the advanced audio properties and look at the track grid for every source, then to open the file in MediaInfo and count how many audio tracks it has, and then to try each track in turn in VLC's audio track menu.

Fourth Layer: Encoder and Writing

This is the layer the MediaInfo track-length comparison tests. The audio encoder compresses the sound and writes it into the container, and if the encoder is overloaded, if the audio buffer fills up, or if the disk cannot keep up with the write, the audio track can stop being written while the video track continues. The check for this layer is the comparison of the audio track length against the video track length, together with the file size against the file length, and whether there was any encoder overload message during the recording, and whether the disk had space. If the audio track is shorter than the video track by exactly the length of the silence at the end, that is the layer.

Fifth Layer: File and Container

This is the layer the player comparison tests. The traditional MP4 format is not crash-safe: if the recording is interrupted by a crash, a power cut, or the program being closed, the file may never be finished properly, and the audio track index may be incomplete, so a player reaches a point in the file and has no more audio to play. MKV is more crash-safe in this respect.

The Seven Diagnostic Steps

Take the steps one after another, in the order given. Each step is one comparison, and each comparison should end with a note of what it showed. Only ever change a single thing in one step: change two things together and the result tells nobody which of them did the work.

StepWhat to checkWhat the result means
Step 1: Baseline and time codeNote the recording tool and version, the file format, the file length, and the exact minute and second at which the sound stopsWithout a time code, "silent halfway through" is a guess, and the rest of the diagnosis has no evidence
Step 2: MediaInfo comparisonOpen the file in MediaInfo and compare the audio track length against the video track length, and count the tracksAudio track shorter than the video track means the encoding or writing layer; audio track the same length means the source, device, or track layer
Step 3: Player comparisonPlay the same file in VLC and in a second player, and try each audio track in VLC's track menuBoth players silent at the same time code means the silence is in the file; VLC with sound in the second half means the reading layer, follow the guide numbered 506
Step 4: Device comparisonCheck Device Manager for the audio device and any yellow warning, check the Windows power settings for USB selective suspend, and check how the recording software binds its sourcesA device that has disappeared, or a power setting that puts it to sleep, is the device layer
Step 5: Re-recording testRecord a test clip of the same length with the same settings, watching the volume meter all the way through, and noting the time code at which it stops, if it stopsMeter stops means the device or source layer; meter keeps moving but the file is silent in the second half means the encoding, writing, or track layer
Step 6: Track comparisonCheck the advanced audio properties track grid for every source, and compare it against the MediaInfo track count and the VLC track menuA source never assigned to the recording track is the track layer
Step 7: Stop lineDecide whether the layers have been separatedIf the seven steps cannot separate the layers, or the file comes with a crash, a power cut, or a file MediaInfo cannot read, stop and bring the records for inspection

Step 1: Record the Baseline and the Time Code

The first step is to pin the fault down in words. Which recording tool made the file — OBS, ShadowPlay, the Xbox Game Bar, or AMD ReLive? What is the file format, and how long is the file? And most importantly, at exactly which minute and second does the sound stop? Play the file in VLC section by section and write the number down. That time code is the one piece of evidence that exists before any setting is changed, and every layer in this guide is checked against it.

Step 2: The MediaInfo Comparison

Open the recording file in MediaInfo and look at what tracks it contains. Compare the audio track length against the video track length, count how many audio tracks there are, and note the codec, the sample rate, and the channel layout. If the audio track is shorter than the video track, the sound stopped being written at that point, and the fault is in the encoding or writing layer — move to step four and step five with that in mind. If the audio track is the same length as the video track, the sound kept being written and what was written was silence, and the fault is in the source, device, or track layer — move to step four and step six. The MediaInfo result has to be written down, because the track lengths are the evidence that separates "stopped writing" from "written silence".

Step 3: The Player Comparison

Play the same file in VLC and in a second player, and compare the two at the same time code. If both players go silent at the same time code, the silence is in the file itself, and the investigation stays with this guide. If VLC still has sound in the second half while the second player is silent, the fault is in how the file is being read, and the road is the one in the guide numbered 506. This step is a comparison, not a repair: it decides whether the silence is in the file or in the reading of the file, and nothing more.

Step 4: The Device Comparison

Open Device Manager and check whether the audio device is still there and whether it carries a yellow warning. Check the Windows power settings for USB selective suspend. Check how the recording software binds its sources — by device name, or by "default device". Write down the result of each check, rather than relying on a feeling that "the device is fine". If the device has disappeared, or if the power setting puts it to sleep, that is the device layer. But the honest limit has to be kept in mind: the sources describe devices disappearing before a recording starts, not during one, so a device fault found here is a hypothesis that has to be confirmed by the re-recording test in step five, not a verdict on its own.

Step 5: The Re-Recording Test

Record a test clip of the same length, with the same settings, and watch the volume meter all the way through. Note the time code at which the meter stops, if it stops, and note whether the scene changes at any point during the test. If the meter stops, the sound stopped reaching the recording software, and the fault is in the device or source layer — go back to step four with that time code in hand. If the meter keeps moving but the test file is silent in the second half, the sound reached the software and was lost at the step where it is written into the file — the encoding, writing, or track layer — and the next step is the track comparison.

Step 6: The Track Comparison

Open the recording software's advanced audio properties and look at the track grid for every source: which tracks each source is ticked for. Then compare that grid against the tracks the recording output is set to record, and against the MediaInfo track count and the VLC track menu results from step two and step three. A source that was never assigned to the recording track is the track layer. A source assigned only to a second track, while the player only plays the first, is also the track layer. This step is a comparison, not a recommended setting: find which track is missing first, and only then is the layer known.

Step 7: Stop Line

If all seven steps have been run and the layers still cannot be separated, or if the file comes with a crash, a power cut during recording, or a file MediaInfo cannot read at all, stop testing and bring the full record for inspection.

Six Misconceptions That Mislead This Diagnosis

1. "It had sound at the start, so the whole file must have sound."

Wrong. Sound at the start only proves the chain was working when the recording began. A device can fall asleep partway through, an encoder can stop writing partway through, a scene can change partway through. The whole file has to be played back before anyone can say the whole file has sound.

2. "The volume meter was moving, so the second half must have been recorded."

Wrong. The meter only says the sound reached the recording software at that point. It does not prove the encoder wrote it into the file, and it does not prove the source was assigned to the recording track. The meter is evidence from the middle of the chain, not evidence about the file.

3. "Updating the driver will fix it."

Wrong. Updating a driver is a variable, not a repair. A driver update can itself trigger Windows to re-enumerate audio devices and give them new identifiers, which is the very mechanism that makes recording software lose a device. The sources disagree on this point, and the disagreement is written out in the section below: some guides list updating the driver as a routine step, while others say a Windows update is itself what triggers re-enumeration and makes a device disappear. Before any update, write down the current settings and the current driver version, and change only one thing at a time.

4. "A mismatched sample rate made it go silent halfway through."

Wrong. A mismatched sample rate causes gradual drift: the sound gets further and further out of sync as the file goes on, but it does not suddenly go silent. Sudden silence is a switch-like event — sound at one time code, nothing after it. Writing drift as the cause of sudden silence is a mechanism error. Drift belongs with the guide numbered 504, where the sound is present but in the wrong place in time. The guide numbered 496 covers a separate drift shape — variable frame rate drift — not sample-rate drift. This guide covers silence, not drift.

5. "VLC plays the second half, so the file is fine."

Wrong. VLC plays only the first track by default. Sound in the second half may be coming from a different track entirely, and the editing program may not be reading that track at all. VLC's own decoders are built in, while editing programs often rely on system decoders, so "VLC plays it" and "the file is fine" are two different statements. Use MediaInfo to count the tracks, and try each track in turn in VLC's track menu, before saying anything about the file.

6. "Switching to MKV, or changing the format, will fix it."

Wrong. Changing the format is a comparison test — it shows whether the problem follows the container — not a repair. Switching to MKV before the layer has been separated only changes the comparison material. MKV is more crash-safe than the traditional MP4 format, which is a real difference, but crash safety is about a recording interrupted by a crash, not about a device falling asleep or an encoder stopping mid-write. Do not present a format change as a universal fix.

Comparison Table: What Each Result Points At

ObservationLayer indicatedNext test
Sound present at the start, silent after one time code, picture normal throughoutNot yet separated — time code comparison firstTime code comparison, step 1
MediaInfo audio track shorter than the video trackEncoder or writing layerRe-recording test, step 5, read against the MediaInfo comparison in step 2
MediaInfo audio track the same length as the video track, second half silentSource, device, or track layerRe-recording test, step 5
Volume meter stops during the re-recorded testDevice or source layerDevice comparison, step 4
Volume meter keeps moving but the test file is silent in the second halfEncoding, writing, or track layerTrack comparison, step 6
Both players silent at the same time codeThe silence is in the fileContinue with this guide
VLC has sound in the second half, the second player does notReading layerFollow the guide numbered 506
Audio device missing in Device Manager, or a yellow warningDevice layerDevice comparison, step 4
Scene change time matches the silence time codeSource layerSource comparison, step 5
Source never assigned to the recording trackTrack layerTrack comparison, step 6
File cannot be read by MediaInfo, or the recording was interrupted by a crashFile corruption roadStop and bring the record for inspection

What the Sources Disagree About

The sources for this guide do not all say the same thing, and the disagreements have to be written out rather than settled silently.

Which layer should be suspected first? A guide on umatechnology.org and a guide on geekchamp.com put track assignment first: a meter that moves while the recording is silent points at a source that was never assigned to the recording track. A guide on obscountdown.com and a thread on the OBS forums put device re-enumeration first: a Windows update gives a USB audio device a new internal identifier, and recording software bound to the old identifier can no longer find it. A Microsoft Q&A thread puts power management first: Windows power management puts USB devices into a low-power state, and the device stops producing sound. No source says which one is the leading cause for this shape, and no source directly documents this shape at all.

Is updating the driver a repair or a cause? Guides in the geekchamp.com style list updating the driver as a routine step. Guides in the obscountdown.com style, and the OBS forums thread, say a Windows update is itself what triggers re-enumeration and makes a device disappear. Both are right, because the cause is different in each case: updating helps when a driver is damaged or outdated, and updating harms when the update itself moves an endpoint or gives a device a new identifier.

How much does "VLC plays it" prove? Several sources treat a file that plays in VLC as useful evidence. But VLC plays only the first track by default, as the multi-track sources in the guide numbered 506 show, and VLC uses its own built-in decoders while editing programs often rely on system decoders. "VLC plays the second half" proves that the first track plays in VLC. It does not prove the file is fine, and it does not prove the other tracks exist. The MediaInfo track count and the VLC track menu have to be read together before anything is said about the file.

Boundary: Shapes This Guide Does Not Cover

SituationWhich guide to followWhy it is not this guide
Recording file silent from the first second to the lastThe guide numbered 503, recording file silent, picture fineThe premise is silence from the start, not silence that begins partway through
Recording file has sound, but it crackles and breaks upThe guide numbered 311, game audio crackling and breaking upThe sound is still there, only degraded
Recording file sound drifts further out of sync as the file goes onThe guide numbered 496, variable frame rate driftThe audio track exists, it is in the wrong place in time, it is not silent
Finished edit sound present but changed pitch or speedThe guide numbered 504, sample rate roadThe sound is present throughout, only its pitch or speed is wrong
Recording file plays with sound in VLC but shows no audio track in the editorThe guide numbered 506, track mapping, codec and container roadThe file has an audio track, only the editor does not read it; in this guide the second half of the file is genuinely silent
No sound can be heard at all, live or recordedThe guide numbered 170, HDMI and DisplayPort playback chainThe premise is that no sound can be heard anywhere, not that a recording file goes silent
Recording or streaming stutters, encoder overloaded, frames droppedThe guide numbered 171, rendering, encoding and network dropped framesThe problem is stuttering, not silence
Recording file has sound but no pictureThe guide numbered 500, finished export video track roadA video track problem, not an audio track problem
Computer crashes or loses power during recording, file cannot be playedFile corruption road, also noted at the boundary of the guides numbered 503 and 500The file structure is damaged, not a settings problem
Overlay will not open, recording button cannot be pressedThe guide numbered 460, overlay switch, shortcut and permissions roadThe recording never started

When to Stop Testing

Stop testing and bring the records for inspection if any of the following apply. All seven steps have been run and the layers still cannot be separated. The recording was interrupted by a crash, a power cut, or the program being closed, and MediaInfo cannot read the file at all — that is the corruption road, not a settings problem. The file is silent from the first second, which is a different shape from the mid-file silence this guide covers, and belongs with the guide numbered 503. The sound is there but coming through distorted — crackling and breaking up when the machine is under load — and that shape belongs with the guide numbered 311. The sound drifts out of sync rather than disappearing, which belongs with the guide numbered 496. The overlay itself will not open, which belongs with the guide numbered 460.

What to Bring for Inspection

If you can bring the machine itself, do; if not, bring the written record instead. What counts is the figures, not the conclusions: which recording tool and version, which output settings, which file format, how long the file is, and the exact time code the sound stops at. Bring the MediaInfo reading as well — how many audio tracks there are, each track's length next to the video track's length, which codec, and which sample rate — together with the outcome of every comparison that was run, naming the step each one came from: how the two players compared, how the device compared, what the re-recorded test clip showed, and how the tracks compared. Inspection and repair pricing is published in three tiers, HK$400–2,800, and the tier that applies depends on what the inspection finds; no price is quoted before the inspection, because the fault has not been located yet. The shop is at Room 1209, Dali Plaza, Shek Mun. The telephone number is 2186 8010, and the WhatsApp number is 6558 6806. There is no shop in Mong Kok, and there is no on-site service.

Frequently Asked Questions

1. My recording starts with sound and goes silent halfway through. Is the graphics card failing?

Not usually. A normal picture with sound that disappears partway through points at the audio chain, not at the card's rendering. The first step is the time code comparison, then the MediaInfo comparison, to see whether the audio track is shorter than the video track or the same length but silent in the second half.

2. How do I find the exact time the sound stops, and why does it matter?

Play the file in VLC section by section and write down the minute and second at which the sound disappears. That time code is itself evidence. Every layer in this guide is checked against it: if the silence begins at the same time as a scene change, that points at the source layer; if it begins at the same time as a Windows power event or a device re-enumeration, that points at the device layer; if the audio track in MediaInfo is shorter than the video track by exactly the amount of silence at the end, that points at the encoding or writing layer.

3. MediaInfo says my audio track is shorter than the video track. What does that mean?

It means the sound stopped being written at that point, and the fault is in the encoding or writing layer. The encoder compresses the sound and writes it into the container, and if the encoder is overloaded, if the audio buffer fills up, or if the disk cannot keep up with the write, the audio track can stop being written while the video track continues.

4. MediaInfo says my audio track is the same length as the video track, but the second half is silent. What does that mean?

It means the sound kept being written, and what was written was silence. That is a different kind of silence from a track that stopped being written, and it points at a different set of layers: the source layer, the device layer, or the track layer. The next step is the re-recording test — record a test clip of the same length with the same settings and watch the volume meter all the way through — to see whether the meter stops, which points at the device or source layer, or keeps moving, which points at the encoding or track layer.

5. I do not remember whether the volume meter was moving when the sound stopped. Can I still diagnose it?

Yes, but not from memory. Whether the meter was moving during the original recording cannot be recovered afterwards, and this guide never assumes the reader remembers it. Record a test clip of the same length with the same settings and watch the meter all the way through, noting the time code at which it stops, if it stops. If the meter stops, the sound stopped reaching the recording software. If the meter keeps moving but the test file is silent in the second half, the sound reached the software and was lost at the step where it is written into the file.

6. Will updating the driver fix a recording that goes silent halfway through?

Not as a general rule, and it can make things worse. Updating a driver is a variable, not a repair. A driver update can itself trigger Windows to re-enumerate audio devices and give them new identifiers, which is the very mechanism that makes recording software lose a device. Before any update, write down the current settings and the current driver version, and change only one thing at a time. The sources disagree on this point — some guides list updating the driver as a routine step, others say an update is itself what triggers the device to disappear. Either way, write down the current settings first and change only one thing at a time.

7. Could a mismatched sample rate cause the sound to stop halfway through?

No. A mismatched sample rate causes gradual drift: the sound gets further and further out of sync as the file goes on, but it does not suddenly go silent. Sudden silence is a switch-like event — sound at one time code, nothing after it. Writing drift as the cause of sudden silence is a mechanism error.

8. VLC plays the second half of my file with sound. Does that mean the file is fine?

No. VLC plays only the first track by default, so sound in the second half may be coming from a different track entirely, and the editing program may not be reading that track at all. Use MediaInfo to count how many audio tracks the file has, and try each track in turn in VLC's audio track menu, before saying anything about the file.

9. Should I switch from MP4 to MKV?

Not first, and not as a fix by itself. Changing the format is a comparison test — it shows whether the problem follows the container — not a repair. MKV is more crash-safe than the traditional MP4 format, which matters when a recording is interrupted by a crash, but crash safety is not the same as a device falling asleep or an encoder stopping mid-write.

10. When should I stop testing and bring it in for inspection?

Stop when all seven steps have been run and the layers still cannot be separated, or when the file comes with a crash, a power cut during recording, or a file MediaInfo cannot read at all. Bring the full record for inspection. Inspection and repair pricing is published in three tiers, HK$400–2,800, and the tier that applies depends on what the inspection finds. The shop is at Room 1209, Dali Plaza, Shek Mun, telephone 2186 8010, WhatsApp 6558 6806.

Sources

Related articles

Related services

If you need Chip-level GPU repair, see these service pages: