Online AI mastering with an analysis-driven mid/side chain

Upload an unmastered mix and Unveil's Mastering tool analyzes it, builds a mastering chain from what it finds, and renders a 24-bit WAV you can compare against the original. The chain works on the mid and side channels separately, then finishes with compression, a true peak limiter and loudness normalization. You need a free Unveil account to run it.

What mastering actually does

Mastering is the last technical step before release. It takes a finished stereo mix and makes small, broad adjustments so the track holds up on different speakers and streaming services. Typical tasks are gentle tonal correction, light compression to even out dynamics, limiting to control peaks, and setting the final loudness.

It is not a second mixing session. Mastering treats the whole stereo file at once, so it cannot lift one buried instrument or remove one harsh element without touching everything around it.

Loudness targets, headroom and true peak

Loudness is measured in LUFS, which weights frequencies the way the ear does. Streaming services normalize playback, and -14 LUFS integrated is a common reference point. Quieter styles such as classical or ambient are often delivered lower, and dense genres are sometimes delivered louder, but pushing loudness past what the music needs mostly costs dynamics.

True peak (dBTP) estimates the highest level the audio reaches after conversion to analog or to a lossy codec. A ceiling of -1.0 dBTP leaves a small margin so the encoded file does not clip. Headroom matters before mastering too: if the mix already hits 0 dBFS from a limiter on the master bus, there is little left for the mastering stage to work with.

Mid/side processing in plain terms

A stereo file can be rewritten as a mid signal (what is common to left and right, usually the vocal, kick and bass) and a side signal (what differs between them, such as wide pads and room sound). Processing them separately lets you treat the center without changing the width.

Unveil's chain uses this idea. The mid channel gets a high-pass between 30 and 50 Hz depending on how much sub rumble is found, plus narrow cuts where vocals are being masked. The side channel gets a 150 Hz high-pass so the sides carry no low end, and a 1.5 dB shelf at 8 kHz for air.

How Unveil's Mastering tool works, step by step

The analysis stage uses Demucs to split the track into drums, bass, other and vocals. Those stems are used only to measure masking between the vocal and the instruments. The audio that is actually processed is your original stereo file.

Sign in with a free Unveil account and open the Mastering tool.
Upload an MP3, WAV, AIFF or FLAC file and start the run. The tool charges 25 ProCoins when it begins.
The backend detects the genre from the audio and picks the loudness target for that genre.
Stem separation and masking analysis find where the vocal competes with instruments, including presence, vowel formant and low-mid regions.
The chain is built and applied: mid and side EQ, compression, optional soft-clip saturation, a limiter at -1.0 dBTP with a 50 ms release, and up to three passes of loudness normalization.
Compare before and after in the A/B player, review the diagnosed issues, and download the 24-bit WAV.

How the compression is chosen

Compression is derived from the measured dynamic range of your file rather than a fixed preset. In the code, a very dynamic track (over 18 dB) gets 3:1, a moderately dynamic track (over 12 dB) gets 2:1, a track that is already compressed but has a buried vocal gets 1.5:1, and a track that is already controlled gets 1.3:1. Soft-clip saturation is added only when the dynamic range is above 10 dB.

Prepare your file

Upload the mix before any master-bus limiter, or at least with generous headroom.
Bounce at the highest quality your DAW offers. WAV, AIFF or FLAC avoids another round of lossy encoding.
Fix balance problems in the mix first. Mastering cannot rebalance individual instruments.
Check the mix in mono. Phase problems and a wobbly low end are mix issues, not mastering issues.

Honest limits and common mistakes

The EQ moves in this tool are driven by vocal masking, so a track with no vocal usually gets few corrective cuts and mostly receives the high-pass, side processing, compression and limiting. The tool does not repair clipping, bad recordings, or an unbalanced mix. A genre is detected automatically, so a track that sits between styles may receive a target you would not have chosen. Always listen to the A/B comparison before accepting the result.

Common mistakes are uploading an already limited master, chasing loudness at the expense of punch, and judging a master only on headphones. Check the result on a second system as well.

What you get

Separates stems only to diagnose where vocals and instruments mask each other, then masters your original stereo file.
Applies a mid/side chain: a high-pass and surgical cuts on the mid channel, plus air and a high-pass on the side channel.
Sets compression from your track's measured dynamic range, then limits at -1.0 dBTP and normalizes loudness to a genre target.
Gives you an A/B player, before and after LUFS, true peak and dynamic range readouts, and a 24-bit WAV download.

Questions

What file do I upload and what do I get back?

You upload an MP3, WAV, AIFF or FLAC file, and unmastered mixes work best. You get a mastered 24-bit WAV, an A/B player to compare before and after, a list of masking issues found, and the exact chain settings that were applied.

What loudness does it aim for?

The tool detects the genre automatically and uses that genre's target. Most genres in the code target -14 LUFS, Cinematic and Jazz target -16 LUFS, and Classical and Ambient target -18 LUFS. The limiter ceiling is -1.0 dBTP.

Do I need an account, and what does it cost?

Yes. A free Unveil account is required, and signed-out visitors are sent to sign-up. A run costs 25 ProCoins.

Analyze your mix — free