Audio that sounds underwater is almost always audio that has lost its high frequencies, picked up excessive reverb, or both. The result is that hollow, muffled quality where voices seem to come from behind a wall or inside a swimming pool. The good news is that the cause is usually identifiable, and most cases can be improved or fully corrected without expensive tools. Here are seven practical fixes, ordered from the simplest physical checks to more involved software solutions.
1. Check for Physical Obstructions on the Microphone
Before you touch any software, look at the microphone itself. The most common physical culprit is something covering or partially blocking the mic capsule. Phone cases with thick edges, lint in a laptop’s mic port, a windscreen that has gotten wet, or even a finger resting too close to the mic hole can all choke off high-frequency sound and create that classic underwater effect.
Foam windscreens deserve special attention. A dry foam windscreen barely affects sound quality, but a wet one behaves very differently. Testing by DPA Microphones found that a soaked windscreen creates a resonance around 3–4 kHz and significantly attenuates higher frequencies above 5 kHz, producing an effect similar to cupping a vocal microphone with your hand. Even after squeezing out the water and letting it air dry for half an hour, the windscreen still did not perform like a fully dry one.1DPA Microphones. The audio consequences of using wind, rain and virus protection on microphones If you have been recording outdoors in damp conditions or your foam cover has seen better days, swap it out and see if that alone resolves the problem.
For phone and laptop users, the fix is even simpler: remove your case, clean out any debris from the mic opening with a dry brush, and test again. You would be surprised how often the “audio problem” turns out to be a physical one.
2. Fix Sample Rate and Codec Mismatches
If audio sounds fine on one device but underwater on another, or if it sounded fine before exporting but muffled afterward, you may be dealing with a digital mismatch. Two common scenarios cause this: a sample rate conflict and overly aggressive lossy compression.
Sample rate mismatches happen when your recording software expects one rate (say 48 kHz) but the audio interface or system is feeding it another (like 44.1 kHz), or vice versa. The result can be audio that plays back at a slightly wrong speed or, worse, gets resampled poorly, smearing high-frequency detail. Check that your operating system’s audio settings, your recording software, and your audio interface are all set to the same sample rate. On Windows, this means checking both the “Sound” control panel and the settings inside your DAW. On Mac, open Audio MIDI Setup and confirm the rates match. Mismatches here are one of the most overlooked causes of muffled playback.
Lossy compression is the other digital offender. Formats like MP3, AAC, and Ogg Vorbis work by deleting parts of the audio signal that a psychoacoustic model predicts you are least likely to notice. At reasonable bitrates this is barely perceptible, but at low bitrates the codec cuts deeper, removing spectral content and leaving gaps in the frequency spectrum.2arXiv. Robust lossy audio compression identification If your audio has been through multiple rounds of lossy encoding (exported to MP3, imported somewhere else, exported again), each pass strips away more high-frequency information. The fix is to always work from the highest-quality source file you have. Export to a lossless format like WAV or FLAC for editing, and only convert to a lossy format as the final step.
3. Roll Back Overzealous Noise Reduction
Noise reduction is one of the most common causes of artificial underwater sound, and it is entirely self-inflicted. Every podcaster, streamer, and video editor has been tempted to crank the noise removal slider all the way up to eliminate background hiss or hum. The problem is that aggressive noise reduction does not just remove noise; it removes chunks of the signal you actually want.
Most noise reduction tools work by estimating what the noise sounds like and then subtracting that estimate from the full audio. When the subtraction is too aggressive, it can pull away parts of the speech or music signal that overlap with the noise frequencies. The result is a wavering, watery artifact sometimes called “musical noise” or “birdie artifacts,” where the audio sounds like it is being played through a drainpipe. This happens because the algorithm occasionally subtracts more energy than is actually present at certain frequencies, and the software compensates by zeroing out those bins entirely, leaving unnatural holes in the spectrum.3ResearchGate. Noise Removal in Speech Processing Using Spectral Subtraction
The fix is restraint. In tools like Audacity, Adobe Podcast, iZotope RX, or OBS’s built-in noise suppression, dial the reduction back until you can hear a little bit of the original noise returning. A small amount of consistent background hiss is far less distracting than the robotic, underwater warble of over-processed audio. If you are using a real-time noise gate or suppressor (common in streaming and video calls), reduce the suppression level by a few dB and test with a friend. You want the lightest touch that makes the noise acceptable, not inaudible.
4. Tame Excessive Reverberation
Recording in a large, bare room or a tiled bathroom produces a wash of reflections that blur speech and make everything sound distant and echoey. That reverberant tail is one of the things people instinctively describe as “underwater,” because it smears the clarity of consonants and transients in much the same way that losing high frequencies does.
Research on how reverberation affects speech perception shows the effect is substantial. In controlled listening experiments, reverberation raised the threshold at which listeners could understand speech by roughly 5 dB on average, meaning listeners needed the target voice to be noticeably louder relative to any competing sounds just to achieve the same comprehension.4PubMed Central. Effects of reverberation and binaural sensitivity on spatial release from masking In everyday terms, reverb does not just sound bad; it actively degrades how well people can understand what is being said.
If you are dealing with reverb in an already-recorded file, software de-reverb tools can help. iZotope RX, Adobe Podcast’s “Enhance Speech” feature, and free tools like the Clarity Vx plugin all attempt to separate the direct signal from the reflected sound. They work best on voice recordings and tend to struggle more with music. Start with mild settings; pushing de-reverb too hard introduces its own artifacts, which can sound, ironically, underwater.
The better fix is prevention. If you record in a room that sounds echoey when you clap your hands, add soft absorption: hang blankets on the walls behind and to the sides of the microphone, place a thick rug on the floor, or record inside a closet full of clothes. Even a pillow propped up behind the microphone makes a measurable difference. Professional vocal booths are not magic; they are just small, heavily padded spaces, and you can approximate the effect cheaply.
5. Use EQ to Restore Lost High Frequencies
When the underwater quality comes from missing treble rather than added reverb, equalization is the most direct software fix. The idea is simple: boost the frequency ranges that have been attenuated. For voice, the clarity and presence of consonants live roughly between 2 kHz and 8 kHz, while “air” and sibilance sit above 8 kHz. For music, lost sparkle in cymbals and acoustic guitars also lives in those upper ranges.
In any EQ plugin or app, try a broad shelf boost starting around 3 kHz, lifted by 3–6 dB. Listen as you increase the boost: at some point the muffled quality will start to clear up, and at some further point it will start sounding harsh and tinny. Back off to the sweet spot in between. If the audio also sounds boomy or tubby (another flavor of “underwater”), a gentle cut in the 200–400 Hz range by 2–4 dB can clean up the low-mid mud.
EQ cannot create detail that was never captured or that was destroyed by compression, though. It amplifies whatever is already in the signal at those frequencies, including any noise. If boosting the highs makes the audio brighter but also hissier, you may need to pair the EQ with gentle noise reduction. Think of EQ as a tonal correction tool: it can rebalance what is there, but it cannot invent what is missing.
6. Try AI-Powered Spectral Restoration
When EQ is not enough because the high-frequency content truly is not there anymore, a newer category of tools can actually generate the missing frequencies using machine learning. These bandwidth extension models analyze the low-frequency content that remains and predict what the high frequencies should sound like, then blend the two together.
Recent work in this area has shown strong results. One approach based on a neural vocoder backbone can extend audio from as low as 8 kHz up to 48 kHz, effectively filling in the entire upper spectrum that was lost to compression or low-quality recording, and it runs fast enough to process audio in real time on a standard multi-core CPU.5arXiv. Fast and Flexible Audio Bandwidth Extension via Vocos Consumer-facing versions of this technology are already built into products like Adobe Podcast’s Enhance tool and Descript’s Studio Sound, which apply bandwidth extension alongside de-reverb and noise reduction in a single pass.
These tools are not perfect. They work best on speech and can sometimes introduce a slightly synthetic quality, especially on music or complex sound effects. They also cannot recover information that is critical for identification, like the exact timbre of a specific instrument. But for podcast episodes, interview recordings, phone calls, and video narration, they can transform a genuinely muffled recording into something that sounds remarkably clean. If you have old recordings from phone interviews or low-bitrate archives, this is probably the most powerful fix available today.
7. Re-Record with Better Mic Technique
Sometimes the best fix is accepting that a recording cannot be fully rescued and capturing it again properly. If re-recording is an option, getting the technique right at the source prevents every problem discussed above and produces results that no amount of post-processing can match.
Microphone distance is the single biggest factor. Speaking too far from the mic means the signal is weak relative to room reflections and background noise, and any gain you add later amplifies both the reverb and the noise floor, producing that hollow, distant quality. For most USB and lavalier microphones, keep your mouth about 15–20 centimeters (roughly a hand’s span) from the capsule. Closer than that and you risk plosive pops; farther and you start losing the direct-to-reflected sound ratio that keeps audio crisp.
If you are recording on a phone or laptop, face the built-in mic directly and stay within arm’s reach. Phone microphones are tiny and omnidirectional, so they pick up a lot of room sound. Holding the phone in your hand near chest height and speaking toward it, rather than leaving it on a table across the room, makes a dramatic difference.
Finally, do a test recording of ten seconds and listen back with headphones before committing to a full session. Headphones reveal problems that speakers hide, especially the subtle muffled quality that leads to the “underwater” complaint in the first place. If the test clip sounds clear and present in headphones, you are in good shape.
When Multiple Problems Stack Up
In practice, underwater audio rarely has a single cause. A common scenario: someone records a video call in a reverberant room, the call software applies aggressive noise suppression, and the resulting file gets exported at a low MP3 bitrate. That audio has been hit by reverb, by algorithmic noise removal artifacts, and by lossy compression all at once. Each layer strips away a bit more high-frequency detail, and by the time the file reaches the listener, it sounds like it was recorded in a fish tank.
When you are stacking fixes, order matters. Start by addressing the most destructive problem first. If the audio is heavily reverberant, run a de-reverb pass before doing anything else, because reverb tails confuse noise reduction algorithms and make EQ boosts sound ringy. After de-reverb, apply gentle noise reduction if needed. Then use EQ or spectral restoration to bring back brightness. Reversing that order, say boosting highs before removing reverb, tends to amplify the reverb tails and make the problem worse.
The same principle applies to AI-powered all-in-one tools. Adobe Podcast’s Enhance and similar products apply de-reverb, noise reduction, and spectral restoration in a carefully tuned sequence internally. Running your audio through one of these tools and then also manually applying heavy EQ and separate noise reduction on top usually does more harm than good. Pick a workflow: either use the all-in-one tool and trust its pipeline, or do each step manually with full control. Mixing both approaches tends to produce a new set of artifacts.
Why Phone Calls and Video Meetings Sound Especially Bad
If you have noticed that underwater audio seems to plague Zoom calls, phone recordings, and voice messages more than anything else, there is a reason. Real-time communication software is optimized for bandwidth efficiency and latency, not fidelity. Most voice-over-IP systems transmit audio at sample rates between 8 and 16 kHz and bitrates that would make an audiophile wince. At 8 kHz sampling, the highest frequency that can be represented is 4 kHz, which cuts off everything above the fundamental range of speech. Consonants like “s,” “t,” and “f” rely on energy well above 4 kHz, so they get blurred or lost entirely.
On top of the bandwidth limitation, these systems apply their own noise suppression and automatic gain control in real time. The noise suppression is tuned to keep the call intelligible in noisy environments like coffee shops and open offices, which means it is often more aggressive than what you would choose for a clean recording. The combination of narrow bandwidth, aggressive suppression, and variable network conditions (which cause the codec to drop quality further during packet loss) is why so many call recordings sound like they were made inside a jar.
If you need good-quality audio from a remote conversation, the best approach is to have each participant record locally on their own device using a proper recording app, then combine the files afterward. Tools like Riverside, Zencastr, and SquadCast are built for exactly this workflow. Each participant captures full-bandwidth audio on their own machine, so you get the quality of a local recording with the convenience of a remote conversation. The internet connection only carries a low-quality monitoring stream; the final audio never touches the VoIP codec at all.
Common Misconceptions About Muffled Audio
A persistent myth is that buying a more expensive microphone will fix underwater sound. In most cases, microphone quality is not the bottleneck. A thirty-dollar USB mic in a treated room, positioned correctly, will produce cleaner audio than a five-hundred-dollar condenser in an echoey kitchen placed three feet away. The room, the distance, and the recording chain matter more than the capsule itself for the kinds of problems that make audio sound muffled.
Another misconception is that converting a low-quality file to a higher-quality format will improve it. Saving a 64 kbps MP3 as a 24-bit WAV gives you a larger file, not a better one. The information that was discarded during the original lossy compression is gone. The WAV is just a lossless container holding the same degraded content. This is where bandwidth extension tools genuinely add value, because they attempt to regenerate the missing spectral content rather than just changing the container.
People also tend to blame their DAW or editing software when the real culprit is the operating system’s audio pipeline. Windows, in particular, sometimes applies its own “enhancements” (spatial sound, loudness equalization, voice clarity modes) to audio playback or recording without the user realizing it. These enhancements can introduce phase shifts, frequency filtering, and dynamic processing that produce a muffled or watery character. If your audio suddenly sounds off and you have not changed your setup, check your system sound settings and disable any enhancements before troubleshooting inside your editing software. On Windows 11, look under Settings, then System, then Sound, then your device’s properties, and turn off any audio enhancements listed there. On Mac, Audio MIDI Setup is the equivalent checkpoint.