The Vault's Artist Tools

LUFS is perceived loudness — how loud a track sounds across its whole length, not how high the waveform gets. It's always negative, because 0 LUFS is digital full scale. -8 LUFS is louder than -14, which is louder than -20.

The headline figure is integrated: one number for the whole track, with silence and near-silence gated out so a quiet intro or a long tail doesn't drag it down. It's the same figure a streaming service computes when it decides how to play you back.

Every major service normalises playback so listeners don't have to ride the volume between tracks. Each has a target, and anything louder than that target is turned down to it on playback.

Master to -8 against a -14 target and the service plays you 6 dB quieter than you made it. The extra level buys nothing — and the dynamics you gave up to get there are still gone. Loudness above the target is a cost with no upside.

The Turned down by column is exactly that, per service:

• -14 LUFS — Spotify, YouTube, Tidal, Amazon Music, Pandora
• -15 LUFS — Deezer
• -16 LUFS — Apple Music

A track quieter than the target is not turned up. 0.0 dB means the service leaves it alone — not that the level is ideal.

True peak is the highest point the waveform actually reaches, including between the samples. An ordinary peak meter only reads the samples themselves and misses the overshoots in between, so a master that reads a safe 0.0 dBFS can still be over the line.

That shows up after encoding. MP3, AAC and Opus rebuild the waveform on playback, and those inter-sample peaks come back as real level — above 0 they clip, which is the crunch on the loudest hits that wasn't in your export. We flag anything above -1.0dBTP. Put your limiter's ceiling there or below and the problem doesn't exist.

Loudness range — LRA, measured in LU — is how much the level moves across the track. A small number means it barely changes; a large one means it does, like a held-back verse into a chorus that actually lands.

It tells you what LUFS can't: two tracks can both read -14 LUFS while one sits flat all the way through and the other breathes. There's no correct value — it's genre and arrangement. Compare it against references you already trust.

In your browser, once, while the file is being decoded for upload. Nothing is sent anywhere to be analysed — the three numbers are worked out on your machine and stored alongside the version, so opening the panel later is just a lookup.

The measurement follows ITU-R BS.1770-4 / EBU R128, the same standard your DAW's meter and the streaming services use, so it should agree with the meters you already check against.

It sits under the player and processes the audio already streaming there — no second download, no server work, nothing saved. Whatever you change applies to what you're hearing right now.

• Separate — pitch and speed move independently. Slow the track without dropping the pitch, or transpose it without changing the tempo.
• Coupled — pitch follows speed, the way a record does. This is what “slowed” actually is.

The range is ±12 semitones in half-steps, and 0.5× to 1.5× speed.

Use it because problems hide at tempo. At 0.7× you hear what your ear glosses over at full speed — a hat landing a hair late, a reverb tail longer than you thought, a click at an edit, a breath you meant to cut. Transposing moves every note in the mix to a new frequency while your room stays where it is: if a boom follows the music it's in the mix, and if it stays put it's your room. It's an evaluation tool — you're listening from a different angle, not making a version.

Five settings, each an EQ curve approximating what a playback system does to a mix's tone. They're filter curves, not measured impulse responses — no room, no speaker distortion, no cabinet resonance. It's a tone check, and tone is the part that decides whether a mix survives. All five are level-matched, so you're comparing tone and not volume.

• Flat — no filtering. Your monitoring as it is.
• Phone speaker — almost nothing survives below about 500 Hz. That isn't an exaggeration for effect: a phone's micro-speaker has no output below its own resonance, so an octave down it's already 25 dB quieter and two octaves down it's gone. If the track only works with its bottom end, this is where you find out.
• Laptop — bottom gone below roughly 200 Hz, but rolling off rather than falling off a cliff, with the low mids pulled back and presence pushed. How it sounds to someone half-listening at a desk.
• Earbuds — consumer tuning: lifted bass, dipped low mids, and a hard peak around 2.8 kHz where the ear canal already adds gain. Sibilance and vocal harshness get found here.
• Car — cabin gain on the bottom, a door-panel bump at 160 Hz, and the top rolled off. Where low-mid buildup shows itself.

The fold sums left and right into one signal. Plenty of systems do that to you — club rigs, a single phone speaker, some Bluetooth speakers. Summing isn't just narrowing, though: content can vanish outright.

When the same sound sits in both channels out of phase, the two cancel as they're added together and you get silence where the part was. Stereo wideners, a doubled part with one side polarity-flipped, an off-the-shelf “wide” sample — impressive on headphones, all capable of disappearing in mono.

The correlation meter tells you whether that's going to happen before it does:

• Near +1 — the channels are nearly identical. The fold changes almost nothing.
• Around 0 — the channels share nothing. Wide, and it survives the fold, though the level shifts.
• Below 0 — the channels are subtracting. That content is what the fold takes away.

Brief dips below zero are normal and not worth chasing. Sustained is the warning — the panel calls it after about a second and a half held under, so you can hear exactly which section does it. When it flags, switch the fold to mono and listen: if the part disappears or the mix loses its body, narrow the widener, check polarity on the doubled source, or keep the low end mono.

Every control in the audition desk is preview. Reset puts them back to neutral, and leaving the page hands the player back exactly as it was. The stored file — the one your collaborators and listeners get — is the one you uploaded, untouched.