Free Online Audio Mastering — Automatic, In-Browser

This is our internal tool. For years we used it as a script: automatically mastering mixes and songs before release. We also had our own A/B comparison program: which mix is better. We combined them here. Use it.

For us, mastering is technical work: bringing a recording up to spec for publication (Spotify, YouTube, CD — does anyone still make those? — or LP), not fixing mistakes. Everything else is solved earlier, at mixing — in the DAW, while you still have separate tracks. The mix is muddy — find the track that is making it muddy and cut the excess there. The vocal is not warm enough — warm its track. Not enough drums — raise their track. That is not mastering’s job: what comes here is an already mixed stereo mix — the separate tracks are gone, and it is too late to tweak them.

The mix lacks space — make the guitar or drums bigger; do not "widen" the record at mastering. And do not crank synthetic brightness and loudness for effect: the track flares up in the first five seconds, and after ten minutes that "improved" sound is impossible to listen to — your ears just get tired.

You can change the settings and run online mastering again without reuploading the file. Each run is saved as a separate block — compare versions and download any of them.

Every platform has its own loudness and peak requirements. Choose the target — the rest is calculated automatically.

Streaming — a universal preset for all music services at once: Spotify, Apple Music, YouTube Music, Amazon Music, Tidal, Deezer, SoundCloud. Preserves the source sample rate.

YouTube — for video; converts to 48 kHz, the standard for video platforms.

CD — classic 44.1 kHz / 16 bit with dithering; the target loudness is noticeably higher than streaming — a CD normalizes nothing: it plays exactly as loud as you master it.

The tool shows the loudness picture even before processing: Integrated LUFS — the perceived whole-track loudness by ITU-R BS.1770, the number streaming services compare against their normalization reference; True Peak (dBTP) — the reconstructed peaks, including inter-sample ones, which set the limiter ceiling; LRA — how much loudness moves around across the track. After mastering, the same metrics are calculated again — you see the "before/after" without the self-deception of "louder means better". For the full story behind these numbers and every service’s reference levels, see the LUFS meter page — it is also the tool to use when you only need to check loudness without processing.

Half the tool is honest A/B. After mastering, you can instantly switch between the original and the master (and between several master versions) at the same point in the song.

The key point is that comparison loudness is matched. So you do not hear "it got louder" (loud always seems better — that is the trap), but exactly what changed: density, clarity, balance, attack, air. This is how you check that the master opened up the track, not mangled it, and that the song stayed the way you intended.

Make several versions with different settings, compare them with each other, and download the one that sounds right.

The same automatic mastering sounds different on jazz and on metal. Four switches define the character:

The starting values already account for track analysis; everything can be overridden manually in the Advanced block.

Mastering is not "make it louder". It is a careful chain of four steps.

High-pass filter (HPF). Cuts sub-bass junk below 15–50 Hz (the exact frequency depends on genre and density); it sits before the compressor so the compressor does not react to infrasound.

EQ. Five bands for final tonal polish — not "painting", but correction. If the mix is balanced, you can leave EQ alone.

Glue compressor. Glues the mix and adds punch. The threshold is calculated not from random peaks, but from the loudest sections of the track (95th percentile short-term LUFS) — the compressor reacts to choruses and drops, not to a single stick click. Ratio, attack, release and knee are selected from Style + Density + Dynamics; for jazz and classical, it can be turned off to preserve live dynamics.

True-peak limiter. Raises loudness to the target and keeps peaks under the ceiling: over several passes it measures the actual LUFS and adjusts gain until it lands on target within about 0.1 dB; oversampling (4× or 8×) catches inter-sample peaks that would otherwise appear after conversion to MP3/AAC. If the selected loudness is too high for this material, the limiter does not drive the track into overload for the sake of a number — it stops at the closest safe level.

Helps when you have a finished mix that needs to be brought to streaming loudness: steady level, platform standards, tidy glue, and clipping protection. For most tracks going to Spotify and YouTube, that is enough.

Also useful for checking a mix while you work. Working versions are usually quiet, and each one has its own loudness — you cannot honestly compare them with finished tracks on YouTube or Spotify, because loud always seems better. One click — and a working mix becomes a release-level mastered copy: you can put it next to a reference at equal loudness and hear what in the mix actually needs fixing.

Does not replace a human mastering engineer where artistic decisions and fixing mix problems by ear are needed — a major-label album, vinyl, soundtrack. And it does not fix the mix itself: phase problems, dirty low end, buried vocal, skewed instrument balance are fixed earlier, during mixing — in the DAW, while you still have the separate tracks. Automation will not save a bad mix; it will bring a good one up to level.

If the problem is guitar tone, first shape the cabinet sound with the cabinet IR utility, and only then master it.

No server upload: all mastering in the browser runs on your device — we do not see your audio, do not store it, and cannot recover it. Close the tab — and nothing is left anywhere. This is useful when the track is not out yet and you cannot send it outside before release. We collect only anonymous usage stats: which preset was chosen, loudness values before/after, processing time, file name and size — no IP address, User-Agent, cookie or fingerprint. The audio bytes themselves are not sent anywhere.

And no ChatGPT, neural networks, or "black box": the algorithm is deterministic — it measures your track, looks at the platform requirements, and calculates settings by explicit formulas. The same file with the same settings gives exactly the same result every time — unlike AI services, where a repeated run can sound different. All settings are visible and can be overridden manually in the Advanced block.

The engine is open: all the real work — loudness measurement, EQ, compression, true-peak limiting, WAV encoding — is done by FFmpeg (ffmpeg.org, license LGPL-2.1+), compiled to WebAssembly via ffmpeg.wasm (MIT, source: github.com/ffmpegwasm/ffmpeg.wasm). The waveform and player on the page are drawn by our own Canvas code — no third-party libraries. Thanks to the authors of the open projects.

Darwin’s Cat is a rock band from Berlin. We make music and tools for our own work, then open them to everyone. Listen: Darwin’s Cat · Contact: band@darwinscat.com · Support: buy us a beer.