AI Mastering
AI mastering that tells you what it changed
Euphona analyses your mix, sets a loudness and tonal target for the material, and applies a mastering chain you can read back move by move.
Meters one minute per minute — a four-minute track costs four minutes of your allowance.
How it works
What it measures, what it changes
And what stays yours to decide. Every stage is a parameter you can read back and overrule.
- 01
It listens to the whole mix first
Before anything changes, the track is analysed end to end: integrated and short-term loudness over time, true peak, crest factor, spectral balance against the material's genre, stereo width by band, and where the dynamics live. We call that analysis the Track Graph. Everything after this step works from it.
- 02
It picks a target, not a preset
A genre embedding picks the reference curves and loudness target that fit the material. A sparse acoustic record and a dense electronic one don't want the same answer; a fixed preset gives them one anyway. You can override the target, and the platform figures we use are the published ones, not ours.
- 03
It applies a chain, not a black box
Corrective equalisation, multiband dynamics, stereo treatment and a true-peak limiter, in that order, with the parameters derived from the analysis. It's a parametric chain, not a generative model, so it can't invent artefacts that were never in your mix.
- 04
You hear it level-matched, then decide
The before and after are loudness-matched before you compare them; an unmatched A/B only tests which one's louder. Every parameter the chain set is in the report, and you can change the ones you disagree with.
The specifics
What you would ask before trusting it with a record
Formats, limits, and the places it is honestly hardest. Nothing here is a range we have not measured.
- Loudness targets
- Integrated loudness measured to ITU-R BS.1770-5, targeted per destination. The platform figures are each service's published normalisation target, and we check the finished master against the target instead of assuming it got there.
- True peak
- True peak measured with 4× oversampling and ceilinged at −1 dBTP by default. Inter-sample peaks above a platform's ceiling distort inside its encoder, after normalisation, and nobody hears it until the release is out.
- Reference tracks
- Supply a reference and the tonal target comes from it instead of from the genre curves. The reference is analysed, never copied: what's taken is a spectral tilt, not a fingerprint of someone else's record.
- Formats in
- WAV, AIFF, FLAC, ALAC, MP3, AAC, Ogg and Opus, at any sample rate up to 192 kHz and any bit depth up to 32-bit float. Decoding happens in a sandbox with no network and no filesystem beyond the file itself.
- Formats out
- WAV and FLAC at the source rate or a chosen rate, 16- or 24-bit with TPDF dither on reduction, plus MP3 and AAC where you want a reference copy. Sample-rate conversion is done once, at the end of the chain.
- The report
- Every job returns the analysis it started from and every parameter the chain set: the EQ moves in dB by band, the dynamics thresholds and ratios, the limiter's gain reduction over time, and the measured before/after loudness and true peak. Measured values, not intended ones.
Questions
What people actually ask
- Will this make my track louder than everyone else's?
- No — and neither will anything else. Every major streaming service normalises playback by measured loudness, so a master 3 LU louder than another simply gets turned down 3 dB on the way out. The dynamic range you spent to get there doesn't come back, though. That part you keep paying for.
- Can mastering fix a problem in my mix?
- A broad tonal one, sometimes. A mix that's uniformly dull or uniformly thin is one problem affecting everything, and a mastering EQ can address that. A buried vocal it can't help — in a stereo bounce the vocal isn't a separate thing any more. If you can name the element that's wrong, fix it in the mix.
- What does the engine do that a limiter preset does not?
- It measures your material and derives the settings from that measurement instead of applying settings tuned on someone else's. The audible difference is mostly in what it doesn't do: it won't compress a track that already has the dynamics its genre expects. A preset has no way to know that.
- Do I have to accept what it chose?
- No. The target, the reference and every stage of the chain are editable, and the report is there precisely so you can see a decision and disagree with it.
Try it
Put your own track through it
The free plan runs the whole engine on your whole song and plays you the result, matched for loudness against the original. No card. You pay when you want the file — one track at a time, or a plan if you are making a lot of them.
Who it’s for