ToolTip AI

Windows audio and MIDI,
device by device.

Sound problems are blamed on drivers more often than on anything else in Windows, and most of them are a format, a role or a buffer that nobody chose deliberately. Thirteen entries, ordered from what the engine does to you, to what you have to install yourself.

wasapi · exclusive · a2dp · hfp · midi 1.0 · midi 2.0 · dpc

How to find the layer that changed your sound.

Sound on Windows is converted before it is played, and the conversion is the story.

Between an application and a speaker there is an engine with its own format, a set of device roles that decide which stream goes where, and a driver that has its own opinions about block size. Almost nothing in that chain is a choice you made, which is why a fault is usually described as random rather than as belonging to one of four places.

The reliable separation is to ask whether the problem is in the data, the format or the ownership. Wrong data means the source; a format mismatch means the engine or the device; and a device that works in one program and not another is owned by something else, because these ports were never designed to be shared.

  • One engine per output. Everything is resampled to the mix format, so the application's rate is a request, not a guarantee.
  • Roles before cables. A headset has two profiles and a microphone has one privacy switch, and either can produce a fault that looks like hardware.
  • MIDI ports are not shared. The first program to open one is the only one that receives, and the second hears nothing while reporting success.
  • Latency has a formula. Block size divided by rate, times buffers in flight; below what the machine can service it becomes crackle, not speed.

The thirteen places sound and MIDI are actually decided.

Ordered from what Windows does without asking, to what you must install for yourself.

Shared mode: the audio engine that mixes everything into one format
What it is
Windows runs one engine per output device, and the engine owns a mix format: a sample rate and a bit depth that everything else is converted to before it reaches the hardware. The value you see in the Advanced page of the old Sound panel is that format, and every application is resampled to it whether it agrees or not.
What changes
Playing a 44.1 kilohertz file through a device set to 48 kilohertz is normal and inaudible, because the resampler is good. Setting the device to a rate your interface cannot produce gives silence and an error code, and setting it high enough to be honest but too high for the machine gives dropouts that look like a driver fault. The per-application volume slider in the mixer appears only once an application is actually playing, which is why the panel looks empty when you open it to fix it.
Risk
Reversible Changing the format affects every sound on the machine at once, and it is instantly reversible in the same two dropdowns.
Exclusive mode, and the silence that follows it
What it is
An application may ask the engine for the device to itself. If it gets it, no resampling happens and latency drops, which is the reason the option exists. While it holds the device, nothing else is allowed to make sound, including the system notification that would otherwise tell you something happened.
What changes
The symptom people describe as Windows breaking is another program holding the device: a player in exclusive mode, or an application that crashed and left the request standing. The recovery is to close the holder or wait for it to release, and if the machine cannot tell you which program it was, the honest answer is that the audio stack reports only that the device is busy. Some interfaces expose their own control panel that grabs the device for monitoring, which produces the same silence from a source nobody suspects.
Risk
Careful An app that takes the device exclusively also takes the sample rate with it, so the next application you open may find a format it did not expect and fail in a way that looks unrelated.
Why your interface is not in the list at all
What it is
There are four reasons a device does not appear, and they look identical from the dropdown. The device can be disabled in the list of hidden endpoints, it can be present but not set as default for the class the application asks for, it can be enumerated under a different role such as a secondary output, or the operating system's privacy setting for the microphone can be switched off for desktop applications and the interface's inputs disappear together.
What changes
The disabled-device list is reached from the old panel with a right-click, and it is the only place a greyed device can be re-enabled; a disabled playback endpoint can also be the recording half of the same cable. The privacy switch is machine-wide for desktop programs and its state explains a whole class of failures in applications that record, including ones that never mention a microphone in their own interface.
Risk
Reversible Enabling a device you disabled earlier returns the configuration you had. Nothing in this list writes to the hardware.
Bluetooth has two profiles, and switching to the wrong one is automatic
What it is
Stereo music travels over the advanced audio distribution profile at high bitrate and one direction. The headset microphone uses the hands-free profile, which is lower quality, mono, and shared with the earpiece speaker. Windows changes profile when an application opens a recording stream, and changes back when it closes it.
What changes
The effect is a headphone that suddenly sounds like a telephone call while a call, a meeting or a recording application is running, then recovers when it stops. Nothing is broken and no setting fixes it permanently, because the profile is chosen by the use and not by a preference. Disabling the hands-free role of that device in the sound panel keeps stereo audio at the cost of the microphone, which is the correct trade for listening and the wrong one for speaking.
Risk
Careful Turning off the wrong role on a device you also use for calls leaves you with good sound and no input, and the failure looks like a dead microphone.
USB audio: one class driver, endless names, and the power setting that glitches
What it is
A compliant USB audio device is handled by the class driver in Windows, which is why it works the day you plug it in and why the name it shows is whatever the manufacturer wrote in the device string rather than the name on the box. Vendor drivers replace that string with the product's marketing name and add their own control panel and features.
What changes
Two power settings interfere with audio specifically: USB selective suspend, which can park a device between bursts of traffic and produce an audible click or a first second of silence, and the maximum power selection on a USB hub, which limits the current an interface may draw and shows up as a device that enumerates and then vanishes under load. Both live in places that do not mention audio at all, which is why the fault is usually attributed to the driver.
Risk
Reversible Disabling power management on a hub is reversible and affects only that bus. It is one of the few audio fixes that cannot break anything else.
The two MIDI stacks, and why your instrument is in one list and not the other
What it is
Windows still carries the legacy multimedia MIDI API, the one every digital audio workstation has used for decades, which exposes simple input and output ports. Above it sits the newer MIDI subsystem, which understands the updated protocol, device properties, capability queries and per-port endpoints, and which is installed as a separate component rather than shipped enabled.
What changes
An application written against the older API sees only the ports the compatibility layer publishes; an application written for the new stack sees the richer endpoint list. That is the entire explanation for a controller that appears in one program and not another on the same machine, and for a device whose extra features are visible in the new configuration utility but unreachable from the sequencer you actually use.
Risk
Reversible Installing the newer subsystem does not remove the old ports, so a machine that worked before keeps working while you find out which API your software uses.
MIDI shows the port but nothing arrives: the four real causes
What it is
A port that opens successfully and carries no data is not a broken cable. The usual causes are an application that already owns the port because MIDI has no sharing, a 32-bit program looking at the 32-bit view of the port list while your device publishes a 64-bit one, a device running on a driver that presents its ports as a single combined endpoint, and an application launched elevated while the MIDI service or the device was enumerated in the normal session.
What changes
Ownership is the one that surprises people: because the older API has no multiplexing for some device endpoints, the first program to open a port is the only one that receives, and the second gets an open that succeeds and a silent stream. Testing with a small monitor that prints incoming traffic separates a device that is not sending from a port that is already taken, and that test costs less time than any driver reinstall.
Risk
Careful Closing the other holder is the fix, and it is invisible work: nothing in the interface tells you a program you left running an hour ago is reading your keyboard.
There is no built-in cable between applications, and that is a design decision
What it is
Windows has no system router that lets one program's MIDI output become another program's input, and no system device that lets one application record what another is playing. Both need a component that is not part of the operating system: a virtual port driver, a loopback utility, or an application that implements routing itself.
What changes
The consequence is that a workflow copied from a forum answer can be impossible on your machine, because the answer assumed a tool was installed. The newer MIDI subsystem adds loopback and routing endpoints, which makes a set of this kind of routing available without a third-party driver, but it only helps applications that adopt it. Audio loopback is a programming interface rather than a device, so only applications that choose to use it can capture another application's output.
Risk
Expensive Installing a driver that creates virtual devices changes the sound and MIDI lists for every application on the machine, and it requires the privileges the operating system reserves for kernel components.
Latency is arithmetic before it is a driver problem
What it is
Round-trip delay is the buffer size divided by the sample rate, multiplied by the number of buffers the engine keeps in flight. A small buffer means a shorter interval between updates and more work per second for the same audio; a large buffer means comfort for the processor and audible lag between pressing a key and hearing it.
What changes
The number to look at is the block size your interface's own control panel sets, not the slider in the application, because the two can disagree and the hardware wins. Setting it below what the machine can service does not produce a low latency; it produces dropouts, and the dropouts are reported as audio glitches rather than as a failed configuration. Power management and other kernel drivers adding long deferred procedures create the same sound at any buffer size, which is why the same project can glitch on a laptop and not on a desktop.
Risk
Careful Chasing latency downward on a machine already at its limit turns a clean signal into crackling, and the crackle is often blamed on the instrument or the cable.
Mono, stereo and the left channel that is actually a centred mix
What it is
Two flags live in the per-device properties and are switched by accident because both sound plausible. Mono folds the two channels together, which is what you want for a single microphone or a hearing aid and what you never want for an instrument. The Enhancements page holds device-level effects, loudness normalisation and, on some drivers, a spatial audio option that pretends a stereo pair is a surround system.
What changes
With mono enabled, a stereo recording becomes a summed signal and the level of anything panned away from the centre changes in a way that sounds like a fault in the source. Disabling audio enhancements is the first thing to try when a level is wrong, when a recording arrives quieter than it should, or when a bit-percent path is required, and it is a per-device setting rather than a global one. The communications tab holds a further automatic behaviour: Windows lowers every other stream when it decides a call has started.
Risk
Reversible Both flags are obvious once you know they exist, and invisible until then; a mono input can be misread as a broken cable for weeks.
The same interface listed three times, with two of them dead
What it is
Windows remembers a device by the path it was attached to. A unit plugged into a different physical connector is enumerated as a new endpoint while the old record stays in the store, greyed and invisible in the normal list. This is the phantom-device pile, and it grows with every front-panel port, docking station and hub change.
What changes
The old records are shown with View, Show hidden devices in the device manager, and removing them there clears the duplicates from the selection lists. Leaving them is harmless for printing but misleading for audio: an application can default to a disabled endpoint that belongs to a monitor that is no longer connected, and the sound then goes nowhere with no error at all. Removing a device that is currently in use gives an immediate silence, which is at least an honest symptom.
Risk
Careful Deleting the wrong entry on a machine with several identical interfaces means unplugging and rediscovering them in the order you want, so write down which is which first.
When the audio stack itself needs a restart, and what that costs
What it is
Two services own the endpoints: the build service that creates the device endpoints and the engine that hosts them. Restarting them makes every application lose and rediscover the devices, which is the fast way out of a state where a device exists but produces nothing, or where the list has stopped updating.
What changes
The restart is audible: streams drop, an application may keep a dead handle and need closing, and the default device selection can change as endpoints are rebuilt. If the underlying problem was a driver that faults, the services restart themselves in a loop and the machine appears to be losing audio repeatedly, which is the pattern that identifies a driver rather than a setting. Recording what the event log says at the moment of the restart is the difference between cycling services forever and finding the package at fault.
Risk
Expensive Restarting these services during a recording or a call takes the sound out from under whoever is using it, and some applications do not recover the stream without being restarted themselves.
Bit-perfect playback, volume mixing and what the app is really doing
What it is
A player that reports bit-perfect output bypasses the engine's volume control, because changing the level would mean resampling or altering the data. The consequence is not a better measurement, it is that the system volume has no effect while that mode is active, and the only remaining control is the hardware or the player.
What changes
Users describe this as the volume being broken. The check is the exclusive flag in the player's own settings and the format line in the sound panel, which shows the engine's rate and tells you whether the path is being bypassed at all. An application that sets the sample rate of the device to match the file is doing the same thing through the back door, and when it closes, other programs can be left with a format they did not choose and start glitching.
Risk
Reversible Switching the mode back restores the normal path in one click. What is worth remembering is that the two modes are legitimate choices with different failure messages.

Audio faults are reported by the application, but decided three layers below it.

Windows has no screen that says which of the engine, the driver and the mixer produced the silence, so people reinstall the driver, which is the one component that was rarely at fault. The order that works is boring and fast: check the format, check the role, check who else owns the port, and only then suspect the package.

Get ToolTip AI on Microsoft Store
Windows 10 build 19041 or newerWindows 1164-bit onlyNo administrator rights
Part of the guides library.