Nobody reads a spec sheet for fun. But anyone who has bought a phone on paper and regretted it in person knows why the numbers matter more than the marketing line above them.
Browser audio tools deserve the same treatment. A vocal remover page will tell you what it does; the useful information sits in the constraints underneath. File size. Track length. What changes once you sign in. Read those first and you can settle in about a minute whether a tool fits the job, without uploading a thing.
Here is how to read the ones that matter.
The number that decides everything: track length
Length is the constraint people hit first, and the one most often missing from the headline. It applies to a vocal remover and a stem splitter alike.
Take stem separation. An AI stem splitter takes a finished song and produces four files (vocals, drums, bass, and everything else) all starting at the same point, so they stay lined up. The limit worth knowing: a guest session accepts tracks up to five minutes, and signing in raises that to ten.
Five minutes covers most pop songs. It does not cover a live set, a podcast episode, or that eight-minute progressive track you wanted to pull apart. Knowing this before you upload saves you the round trip.
The number people misread: 50 MB
Fifty megabytes sounds generous until you understand what it is measuring.
An MP3 of a four-minute song runs somewhere around 8 to 10 MB. The same song as an uncompressed WAV runs closer to 40. So the cap is comfortable for compressed audio and tight for the format you actually want to feed it, because uncompressed files hold the detail that separation relies on.
The practical read: working from a WAV that sits near the ceiling, cut out just the passage you care about and send that instead of the whole song. Nothing is lost, because a vocal remover only ever works on what you hand it.
Signed in versus not: the line is at download
This is the part worth being precise about, because it is where most tools hide their real pricing.
You can run a vocal remover without an account. Upload, process, preview the result in the browser. Signing in becomes necessary when you want the file on your machine. On the stem splitter it also raises that ten-minute ceiling, adds format choice on individual downloads, and lets you pull all four stems down as one ZIP.
So the free tier is not a demo. It is a full test drive with a gate at the exit. For settling the one question that matters, whether a vocal remover can handle your particular song, the preview is enough. You hear the result before anything costs you.
Two outputs or four: pick before you upload
A vocal remover and a stem splitter look similar from outside and answer different questions.
Run a vocal remover and you get two files back: the singing by itself, and the accompaniment by itself. That is the right call for a backing track, or for hearing what an arrangement is doing once the lead is out of the way.
A stem splitter goes to four: vocals, drums, bass, other. Useful when the thing you care about is the bassline or the groove rather than the singing. The trade is that a four-way split has more chances to leak. Kick drum and bass sit in the same frequency range, so complete separation is not always possible, and the page says so plainly.
One clarification that saves disappointment. “Other instruments” is one file containing guitars, keys, strings, synths, and brass together. It is not a per-instrument split. Nothing on the market hands you the original studio session, and this page is direct about that too: source separation, not the multitrack.
The format list tells you what the tool expects
Supported formats are a signal, not a footnote.
A vocal remover and the stem splitter accept the four common ones: MP3, WAV, M4A, OGG. The note detector, which is the mp3 to midi tool, accepts two more on top of those, FLAC and AAC, and will also record straight from a microphone for up to five minutes.
That last one is the most under-advertised feature in the set. If you want the notes for a riff, you do not need to find a recording of it. Play it, hum it, sing it into the laptop, and convert that. The file you would have gone hunting for never has to exist.
What the numbers do not tell you
Two things, and both are the kind of thing a spec sheet structurally cannot cover.
Whatever you upload caps the quality of everything after it. A file ripped from a video, or an MP3 that has been re-encoded a few times on its way to you, has already lost the high-frequency detail these models use to tell one sound from another. The tool cannot restore what was discarded before it arrived. Two files can both be “under 50 MB, correct format” and produce completely different results.
And running audio through a separator settles nothing about ownership. A stem lifted out of a commercial release is still that release. For practice on your own machine nobody cares; for anything public, the licensing is a separate question that has nothing to do with which tool made the file.
A short checklist before you upload anything
Four questions, about a minute:
How long is the track?
Past five minutes, signing in stops being optional.
What format do you have?
If a WAV exists, use it. If only an MP3 exists, proceed, but judge the output knowing the input was already thinned.
Two outputs or four?
Decide by what you actually want to hear. Reaching for four stems when a vocal remover would have given you the instrumental adds failure points for no benefit.
Does it need to leave the browser?
If you are only checking whether a song separates cleanly, the preview answers that, and you can stop there.
None of this is complicated. It is the habit of reading the constraints before the feature list, the same habit that stops you buying a phone with a battery that cannot survive your commute.