Choose what it should do, then drop your tracks and hear a free preview before you pay. Built for ambient, focus and sleep music, where louder is not better.
What should it do?
Several files are
Drop audio files hereor click to choose · WAV, FLAC, MP3 ·
Stems
Mixer role, level and mute of each stem; double-click a fader to reset it. Changes re-read the analysis a moment later.
Before the stems are added up every step is small and listed above; switch off what you do not want
Your stems stay on the server for an hour after the last mix, so you can change levels and mix again.
Variants pick one to hear it and see its analysis; ★ = the one picked automatically
Stems
Original in the player is by default the stems as exported, just added up (so the player shows everything that was done: mix and mastering); switch to our mix to hear only the mastering
Stem checks what was found in the stems and what was done to them
Checks
Loudness over time originalmasteredshort-term LUFS
Stereo image each dot is a moment of the sound: up = in the middle, sideways = wide; both plots use the same scale
original
mastered
Stereo over time originalmasteredL/R correlation: 1 = mono, 0 = wide, below 0 = out of phase
Loudness change over time mastered minus original, 3 s windows; flat at 0 = the dynamics are the original's
EQ over time
Mid click a map to pin a moment, double-click to follow the playhead again
Side the width of the sound; limited by the mono budget
hover for time, frequency and gain
EQ at 0smid appliedmid, model's requestside appliedside, model's requestside limitdB
Stereo width over time L/R correlation: dashed = original, solid = processed
EQ by segment mean mid gain in dB
Limits derived from this track
Time alignment with the source lag in ms, should stay at 0
Original waveform
Mastered waveform
Average spectrum originalmastereddBFS · 1/6 octave
hover the plot for exact values
Spectrogram · original 20 Hz bottom, 20 kHz top, same colour scale
Spectrogram · mastered
Spectrogram · difference (mastered − original)
blue = mastered is quieter, orange = louder, black = no change (±18 dB scale)
hover a spectrogram for time, frequency and level
Numbers
More measurements dynamics, tone, stereo and what changed against the original
How it works
Say what you want, hear it first
01
UploadWAV or FLAC. Your file is only used to make your preview and master.
02
DescribePick a chip or write it: "wider stereo", "smoother", "less reverb". No knobs.
03
Preview freeA 20 s clip, mastered, with an A/B switch against your original at matched loudness.
04
Pay per trackHappy with it? Pay for that track and download the full-length WAV at your original sample rate.
Made for ambient, focus and sleep music: we match your loudness instead of making everything louder, and keep true peaks under -1 dBFS.
Under the hood
A model is the start, not the master
The listening is done by SonicMaster, an open text-guided mastering model. On its own it changes more than you asked for: it rolls off the highs, drifts out of time, dips in level every 20 seconds and, when it widens, decorrelates the bass so the mix suffers in mono. Most of this service is what we built around it to keep what it does well and take away what it breaks.
A
Your sound stays yoursWe do not play you the model's audio. We measure what it changed, per frequency, in the middle and in the sides, and apply that change to your original as a zero-phase EQ that follows the music. Your timing, dynamics and detail stay; the model's seams do not.
B
Limits come from your trackA change the model makes steadily is kept; one that swings from second to second is mostly dropped. The sides may be raised only as far as your own stereo image allows, so the mix does not fall apart in mono. No fixed settings.
C
The air staysAbove about 4 kHz the sound is your original's. The model takes the top off (on a track we measured: 3 dB at 6 kHz, 21 dB at 17 kHz); we put it back.
D
Fixes to the model itselfIts chunks come back 46 ms short, so a long track drifts and drums double. Its 30 s cross-fades dip the level by about 3 dB. Both are corrected, and the page proves it with a time-alignment and a seam check.
E
Every master is measuredClipping, polarity, clicks, hiss, harshness, tilt, transients, mono loss and loudness over time, each against your original. If something got worse, it says so in plain words instead of hiding it.
F
Stems, mixed with careDrop the stems of a song: we tell bass, drums, vocals and the rest apart, fix polarity and rumble, ride the vocal by at most 3 dB, make room for it, calm uneven stems, and keep the mono sum honest. Every step is a switch, reports what it did, and you can compare with the plain sum of your stems.
All of it runs on our own hardware, and your music is deleted an hour after the result is ready. Thresholds are judgement calls, tested on real tracks and on synthetic faults; we publish what we measured and where it is still thin.