Free Online Mastering — No Signup, No Upload

What mastering actually is (and isn't)

Mastering is the final translation layer between your mix and the systems people hear it on — phone speakers, earbuds, car stereos, club PAs. A mastering engineer balances the overall tone, controls dynamics so the quiet and loud parts both work, brings the level to a competitive loudness for the destination, and ensures the file is technically clean: no clipping, no inter-sample overs, correct format.

What mastering is not: a rescue service for a broken mix. If the vocal is buried or the low end is a mess, mastering can only polish the problem. The classic rule holds — fix it in the mix. That's also the honest limitation of this tool and every automated masterer: it shapes the whole stereo file at once. It cannot turn the vocal up.

LUFS: the only loudness number that matters

For decades engineers measured peaks — the tallest sample in the file. But peaks say nothing about how loud something feels. Two tracks can both peak at 0 dB while one feels twice as loud, because perceived loudness comes from sustained energy, not momentary spikes. The broadcast world solved this with ITU-R BS.1770, the standard behind the LUFS unit (Loudness Units relative to Full Scale) — and it's the measurement this tool implements, not an approximation.

The measurement works in three stages. First, the audio is passed through a K-weighting filter: a high-shelf boost around 1.5 kHz approximating how the head and ears emphasize upper mids, plus a low-cut around 38 Hz, because deep bass contributes less to perceived loudness than its energy suggests. Second, the filtered signal's mean-square energy is computed in 400-millisecond blocks with 75% overlap. Third — and this is the clever part — the blocks are gated: anything below −70 LUFS absolute is ignored (silence shouldn't lower the average), then anything more than 10 LU below the ungated average is also dropped, so long quiet intros don't drag the number down. What remains averages into the Integrated LUFS — one number describing the loudness of the whole track the way a human would experience it.

When you drop a file above, that full gated computation runs on every sample, and the same measurement runs again on the rendered master so the report shows true before/after numbers, not estimates.

How loud should your master be in 2026?

The loudness war is over, and the streaming platforms won it. Every major service now normalizes playback: if your master is louder than their reference level, they simply turn it down. Crushing your dynamics to hit −6 LUFS buys you nothing on Spotify — the listener hears it at −14 anyway, except now it's flat and fatiguing at the same volume as a dynamic master.

• Spotify: −14 LUFS reference, true peak below −1 dBTP (−2 dBTP if you expect heavy transcoding)
• Apple Music: −16 LUFS with Sound Check enabled
• YouTube: approximately −14 LUFS
• Club / DJ play: −8 to −9 LUFS is common — no normalization on a PA system
• Podcasts / spoken word: −16 to −19 LUFS mono-equivalent

Inside the chain: what happens to your audio

The signal path here is a real mastering chain, not a single magic slider. In order: a 24 Hz high-pass filter (4th-order Linkwitz-Riley) removes subsonic rumble that eats limiter headroom without being audible. A three-band EQ applies the tonal corrections from the analysis pass — capped at ±4 dB, because mastering EQ should season, not cook. The signal then splits into three bands at 250 Hz and 4 kHz through Linkwitz-Riley crossovers, and each band gets its own compressor: slower in the lows (so the bass breathes), faster in the highs (so cymbals stay controlled). Multiband compression is what lets a master feel glued without the bass pumping the vocal.

After the bands sum back together, a mid/side width stage opens the stereo image slightly — with the side channel high-passed so the low end stays mono-compatible on club systems and phone speakers. Then a gentle tanh saturation stage — oversampled 4× to prevent aliasing artifacts — adds the harmonic density that reads as 'expensive' on small speakers. The exported master runs through a look-ahead true-peak brickwall limiter (1.5 ms look-ahead, 4× oversampled inter-sample peak detection) and a two-pass loudness normalization, so the file lands on your preset's exact LUFS target with the true peak held below −1.0 dBTP — the same architecture hardware mastering limiters use.

Inter-sample peaks deserve a sentence: a digital file can read 0 dB on every sample yet reconstruct, in the analog converter, a waveform that swings above full scale between samples. That's what 'dBTP' (true peak) measures, why lossy encoders (MP3, AAC) clip masters that 'didn't clip', and why every preset here leaves a −1 dB true-peak ceiling.

What the adaptive analysis actually does (full transparency)

When marketing says 'AI mastering', it's worth asking what the machine actually decides. Here is our entire answer — no black box. When you load a file, the engine measures three things: the average spectrum (a windowed FFT sampled across the whole track), the integrated LUFS, and the crest factor — the gap between peak and average level, which describes how dynamic the material is.

Those measurements map deterministically to settings. If your low band is weak against the reference balance for your chosen target, the low shelf comes up — never more than 4 dB. If the track is very dynamic (crest factor above 16 dB), the compressors lean in with a 3:1 ratio and lower thresholds; if it's already dense, they relax to 2:1 and mostly add glue. Input gain is trimmed so the chain receives a consistent level regardless of how hot your bounce was. Every decision is a rule you just read — adaptive, explainable, and the same every time for the same file. We think that honesty matters more than the word 'neural' in a landing page.

The loudness-matched A/B: the most honest button on this page

Here's the oldest trick in audio sales: make the 'after' louder. Human hearing reliably judges louder as better — better bass, better clarity, better everything — for differences as small as half a decibel. Most mastering plugins and online services exploit this: their bypass button compares your quiet mix against their loud master, and the master 'wins' even when it's actually worse.

The A/B toggle above is gain-compensated: when you switch to the original, it's raised to the same perceived loudness as the master. What you're left comparing is tone, density and balance — the things mastering actually changes. If the master doesn't sound better under matched loudness, don't use it. That's the test every processing decision should survive, and it's the habit that will improve your mixes faster than any plugin purchase.

Headroom: how to bounce your mix for mastering

Whether you use this tool or a human engineer, the input matters. Bounce your mix with the master fader untouched and no limiter on the mix bus — leave 3 to 6 dB of headroom, peaks landing around −6 dBFS. Don't normalize the bounce. Export at the session's native sample rate (44.1 or 48 kHz) in 24-bit WAV; never master from an MP3 if the WAV exists, because lossy artifacts get amplified by the very compression and saturation that make masters sound good.

If your mix already slams into a brickwall limiter at −7 LUFS, there is nothing left for mastering to do — the dynamics are already spent. The analysis card above will tell you this honestly: a crest factor under 9 dB means the file arrived pre-crushed.

When to use a machine, and when to book human ears

An adaptive engine like this is genuinely the right tool for demos, reels content, quick references for A&R submissions, podcast music beds, and checking how your mix translates at streaming loudness. It is consistent, instant, free, and private.

Book a human when the release matters commercially. An engineer hears context a spectrum analyzer cannot: that the sibilance is the singer's diction and not the EQ, that the second chorus needs half a dB the first doesn't, that this genre's audience expects the kick to sit differently. At 12NOTEZ we master on calibrated monitoring with hybrid analog-digital processing from ₹3,999 per song — and the first thing we do is exactly the loudness-matched comparison you just read about. If you've used the tool above, send us that master with your mix; hearing the gap between them tells us precisely what your record needs.

Frequently Asked Questions