You export the instrumental, drop it into your DAW, and the whole thing sounds like it is being played through a wall. The cymbals have no air. The bass guitar has turned into a low hum that sits exactly where the kick drum should be. When the vocal comes back in for the chorus, there is a faint ghost of it still hanging in the stereo field, but it is smeared and phasey rather than clean.
This is the most common complaint people bring to a music separator, and it is also the most misunderstood. The underwater feeling is rarely one single failure. It is usually the sum of three separate problems stacking on top of each other, and they get worse the harder you push the software to work.
Three Different Causes, One Similar Symptom
Before you change a single setting, work out which of these you are actually hearing. They sound similar but need opposite responses.
- Source damage. The file was already loud, clipped, or heavily compressed before it ever reached the separator. There is no clean space left between the instruments to separate.
- Mode mismatch. You used a fine mode on a source that does not have enough clean data to reward the extra effort, or a fast mode on a dense mix that needed more careful handling.
- Mix residue. The separation worked reasonably well, but the leftover vocal bleed or the collapsed low end is what you are actually reacting to — not the separation itself.
Each one has a different fix. Trying to solve all three at once by dragging sliders is how people end up with a worse result than they started with.
What "Underwater" Actually Sounds Like
Vague dissatisfaction is not a diagnosis. Listen for specific signs instead.
- Loss of high-frequency detail. Hi-hats and acoustic guitar strings lose their attack and turn into a soft hiss. This usually points to aggressive processing on a source that was already thin.
- Wobbling or pumping in sustained notes. A held synth pad or a bowed string wavers in volume. That is classic separation residue, where the algorithm is guessing at what belongs to which layer.
- A hollow centre. The middle of the stereo image sounds scooped out, and the bass has no definition. This is often a mix issue rather than a separation issue.
- Vocal ghosting. You can still hear the shape of the singer's phrasing even though the words are gone. This is bleed, and it is expected on dense material.
If you cannot describe the problem in one of these four terms, you are probably comparing the separated result to a memory of the original rather than to the original itself.
A Decision Path You Can Run in Five Minutes
Work through this in order. Stop as soon as the result is good enough for your purpose.
Step 1 — A/B Against the Original, Not Against Your Memory
Play the original file and the separated output back to back at the same volume. If the original already sounds thin, compressed, or noisy, the separator is not the problem. Lower your expectations for that file and move on.
Step 2 — Decide Whether You Need Fast or Fine
Fast processing is the right choice for a quick check, a rough rehearsal reference, or anything you will not publish. Fine processing is worth the extra wait when the source is a clean studio recording and you intend to use the output seriously. If your source is a phone recording from a loud room, fine mode will spend more time producing a similar result — the extra effort cannot recover information that was never captured.
Step 3 — Judge the Balance, Not Just the Separation
This is the step most people skip. Check the vocal and accompaniment levels independently rather than accepting the default balance. A small adjustment to the accompaniment level often fixes the "hollow" feeling, because you have been listening to a level problem and blaming the algorithm.
Step 4 — Set a Stop Condition
Decide in advance what "good enough" means. A backing track for personal practice has a much lower bar than a track you plan to publish. If you find yourself on the fifth round of tweaking, the honest answer is usually that the source file, not the settings, is the limiting factor.
When to Walk Away From a File
Some material simply will not separate well, and recognising that early saves hours. Be cautious with anything recorded through a phone microphone in a reverberant room, heavily limited masters where the dynamic range has been squashed flat, and low-bitrate files where compression artefacts have already blurred the boundaries between instruments.
Results vary with source quality, mixing, noise, and compression. No tool can guarantee identical or perfect separation, and any product claiming otherwise is describing something other than what you are hearing.
What This Means in Practice
If a separated track sounds underwater, resist the urge to reprocess it again and again. Run the file through the four checks above, identify whether the fault is the source, the mode, or the balance, and act on that one thing. On a clean source, the fix is often a balance adjustment. On a rough source, the fix is accepting the limitation and choosing a lower target.
One practical detail worth remembering: extraction, separation, and export all happen on your device. Your source files are not uploaded anywhere, which means you can experiment freely without worrying about what happens to someone else's recording. You still need permission to process and publish whatever you are working with — the tool does not grant you any rights to the music or video itself.
The underwater sound is a signal, not a dead end. Read it correctly and you will know within a few minutes whether to adjust, accept, or abandon the file and find a better one.