How to tell if a WAV was upscaled

Here's the thing about WAV that trips everyone up: a WAV file has no quality of its own. It's a container. It holds raw PCM samples and adds a small header saying how to read them. That's the entire format.

Which means a WAV made from a 96 kbps MP3 is a 50 MB file that sounds like a 96 kbps MP3. It is, in every meaningful sense, worse than the MP3 — same audio, thirty times the disk space. And because people have learned the shorthand "WAV = uncompressed = best," this is the format where bad files hide most comfortably.

Why file size tells you nothing

Uncompressed PCM has a fixed, completely predictable bitrate. Sample rate × bit depth × channels, and that's it:

Every 96 kHz / 24-bit stereo WAV of the same length is exactly the same size, whether the audio inside came from a studio master or a YouTube rip. With FLAC or MP3 you can at least raise an eyebrow at a suspiciously small file. With WAV that signal doesn't exist. You have to look at the audio.

Failure 1: The sample rate was inflated

This is upscaling proper. Someone takes a 44.1 kHz file and resamples it to 96 kHz, and now it's "hi-res." The file doubles in size. No new information enters the building.

It's easy to catch, because sample rate sets a hard ceiling on what a file can contain. At 96 kHz the Nyquist limit is 48 kHz, so a genuine 96 kHz recording will show something above 22 kHz — noise, harmonics, ultrasonic junk from the microphones, whatever. An upsampled file shows nothing. The spectrum stops flat at 22.05 kHz and the entire upper half of the spectrogram is a black void.

When you see that, the verdict is unambiguous: this file was 44.1 kHz, and someone made it bigger.

Failure 2: The bit depth was padded

The companion trick. Converting 16-bit audio to 24-bit appends eight bits to every sample, and those eight bits are zeros. The file is 50% larger and carries precisely as much information as before.

Bit depth controls dynamic range — roughly 6 dB per bit. 16-bit gives you about 96 dB and puts the noise floor around −96 dBFS. Genuine 24-bit gives you about 144 dB in theory, and in practice real recordings land somewhere below −110 dBFS because the analog gear has its own noise.

So: a real 24-bit file has a noise floor well below −96 dBFS. A padded one hits a wall at exactly −96 dBFS, because that's where the original 16-bit source ended. If you can look at the raw samples, the giveaway is even blunter — the bottom eight bits are zero on every single sample.

Failure 3: It came from a lossy file

The worst one, and the most common. Someone had an MP3 or an AAC, decoded it, and saved the result as WAV. The container is now technically lossless — it is faithfully storing every sample of a file that already had its high frequencies deleted.

This looks exactly like a fake FLAC: a hard, ruler-straight cutoff somewhere between 15 and 20 kHz that doesn't move for the whole track. Typical ceilings are ~16 kHz for a 128 kbps source, ~19 kHz for 256, and ~20 kHz for 320.

Note the interaction with failure 1, because it's nastier than either alone: a 96 kHz / 24-bit WAV made from a 128 kbps MP3 is a 33 MB-per-minute file whose audio stops at 16 kHz. Every number in the metadata says "premium." The spectrum says "you have been sold a 128 kbps MP3."

There's a second, subtler signature lossy sources leave behind even when the cutoff looks ambiguous. Lossy encoders process audio in fixed-size blocks, and at low bitrates the transitions between those blocks can leave a faint rhythmic grain in the spectrogram — a regular vertical striping synchronized to the encoder's frame rate rather than to the music. Clean PCM never shows it. You won't always spot it, but when you do, it's as conclusive as the cutoff: blocky texture means a block-based codec has been here.

What a legitimate WAV looks like

Depends where it came from, and it's worth being clear that plenty of honest WAVs are 44.1 kHz — that isn't a defect:

A CD rip should be 44.1 kHz / 16-bit with content up to about 22 kHz and a ragged, music-following top edge.

A genuine hi-res master at 96 kHz should show real content above 22 kHz. It'll usually be low-level and unglamorous — mostly noise and harmonics you can't hear — but it should be there, and it should not stop at a suspiciously round number.

A 24-bit file should have a noise floor meaningfully below −96 dBFS.

And across all three cases, the universal rule applies: every claim in the metadata should be corroborated by something in the audio. Sample rate by content above the old ceiling, bit depth by a lower noise floor, losslessness by the absence of a codec's fingerprints. One honest measurement is worth any number of impressive-looking format strings.

A note on AIFF

Everything here applies identically to AIFF. It's the same uncompressed PCM with a different header — Apple's format rather than Microsoft's — and it can be upsampled, padded and transcoded in exactly the same ways. If you work on a Mac and your "masters" are AIFFs, check them with the same suspicion.

Why people upscale in the first place

It's worth separating the innocent cases from the profitable ones, because you'll meet both.

The innocent cases come from workflows, not fraud. A mixing engineer receives 44.1 kHz stems and bounces the session at 96 kHz because that's the session standard. A DJ's recording software defaults to 24-bit regardless of the source. A distributor's automated pipeline resamples everything to a uniform spec for delivery. In each case the file got bigger without anyone making a quality claim — the upscaling is a side effect, not a sales pitch.

The profitable cases are deliberate. Hi-res download stores charge a premium over CD-quality, and more than one label has delivered upsampled CD masters to meet that demand — when Qobuz audited its catalog it found a meaningful percentage of "hi-res" titles had no content above 22 kHz at all. Secondary marketplaces are worse, because there's no audit at all. And sample-pack and DJ-pool sellers live in an ecosystem where WAV is mandatory, which quietly laundered a generation of lossy source material into "production-ready" files.

The lesson isn't that everyone is lying. It's that the supply chain has no checkpoint between "file meets the technical spec" and "file actually contains the information the spec implies." That gap is exactly where you, a spectrogram, and five seconds come in.

WAV versus FLAC: same audio, different box

A question that comes up constantly: if WAV can hide problems, is FLAC safer? No — safer against nothing. FLAC is the same PCM data with lossless compression applied; decode one and you get back bit-identical samples. Every failure in this article survives conversion in both directions. An upscaled WAV becomes an upscaled FLAC. A FLAC made from an MP3 becomes a WAV made from an MP3.

The differences are practical, not qualitative. FLAC carries proper metadata and checksums, uses half the storage, and its compression ratio happens to be an extra diagnostic signal (lossy-sourced audio compresses suspiciously well — a trick covered in the FLAC guide). WAV's advantages are universality and zero decode overhead, which is why studios and DAWs live on it. Choose the container for the workflow; verify the audio the same way regardless.

Frequently asked questions

Is 32-bit float WAV different? Yes, and it's increasingly common because field recorders and DAWs use it internally. A 32-bit float file captures essentially unclippable dynamic range at the recording stage, which is genuinely useful. But it inherits the same rule as everything else: converting a 16-bit source to 32-bit float adds precision to nothing. Check the spectrum and noise floor, not the format badge.

Can upsampling ever add anything? No — it can only avoid losing. A good resampler changes the sample rate while altering the audio as little as possible; it cannot synthesize harmonics that were never recorded. There are legitimate reasons to upsample (some DACs measure better fed high rates, some plugins alias less at high session rates), but "it became hi-res" is never one of them.

What about DSD and SACD rips? DSD is a different animal — a 1-bit format at 2.8 MHz whose "noise" is deliberately shaped into ultrasonic territory. Converted to PCM (which is how you'll usually receive it, as a FLAC or WAV), genuine DSD-sourced audio shows a characteristically steep ultrasonic noise rise rather than content. It's beyond the cutoff test, but its shape is unmistakable once seen.

My player says the WAV is 9,216 kbps. Isn't that proof of quality? It's proof of arithmetic. 192 kHz × 24 bits × 2 channels is 9,216 kbps by definition, whether the file contains a symphony or a modem recording. PCM's bitrate describes the container's capacity, never the content's quality — which is this article's whole point in one sentence.

The verdict

WAV is a good format for what it's for — editing, archiving, feeding a DAW. It's just a terrible indicator of quality, because it faithfully preserves whatever you hand it, including garbage. The number in the filename tells you how much disk you're using, not how much music you have.

Drop the file into the analyzer and check the measured peak frequency against the Nyquist limit for its stated sample rate. If the ratio is nowhere near 1.0, the file is carrying a lot of empty space.

Terms like Nyquist, PCM and dither explained plainly in the audio glossary.