Audio mastering: online services vs a mastering engineer

The mix is finished. It holds up in headphones, it survives the car, and now there is a last step between it and release. One route takes minutes and runs on an algorithm. The other takes days and runs on a person in a treated room. Both are legitimate. They are good at different things.

What mastering is actually for

Mastering is the last stage before distribution. It takes a finished stereo mix and prepares it to hold together across systems that have nothing in common — earbuds, a car, a phone speaker, a club rig, a streaming platform's encoder.

In practice that means some combination of:

• Tonal correction — broad EQ moves so the spectrum translates rather than collapsing on small speakers or turning harsh on bright ones.
• Dynamic control — compression and limiting to manage the range without flattening the transients that give a record its life.
• Stereo work — tightening or widening the image, and checking that it survives being folded to mono.
• Level targeting — landing the record in a sensible loudness range for where it will be played.
• Deliverable preparation — producing the exact files, sample rates and spacing the destination requires.

What mastering is not is a rescue service. It works on the balance you hand it. A vocal buried in the mix will still be buried, slightly louder.

Where each approach is strong

Automated services analyse the file and apply processing derived from a model. You upload, you get something back quickly, and you can usually preview before committing. Some let you supply a reference track and match toward it, turning a vague instruction into a more concrete target.

A mastering engineer works in a room whose acoustics are known and on monitoring that is calibrated, and brings judgement that is musical rather than statistical. The part that has no automated equivalent is the feedback loop: an engineer can tell you the low end is masked, the vocal sits too bright, or the mix needs another pass before mastering will help at all. Software will process what you gave it without comment.

What each approach gives you, and where it stops

Dimension	Automated service	Mastering engineer
Turnaround	Minutes, at any hour	Days to weeks, subject to a schedule
Cost shape	Subscription or per-track, published on the vendor's site	Quoted per track or per project by the engineer
Consistency	Highly repeatable — the same input gives the same output	Repeatable through the engineer's judgement and notes
Feedback on your mix	None. It processes what it is given	Explicit, and often the most valuable part
Album coherence	Tracks are processed independently	The record is treated as one body of work
Unusual material	Strongest on conventional, dense arrangements	Handles sparse, acoustic and dynamic material on its own terms

Scroll to see all columns

Because service pricing moves and engineers set their own rates, no figures appear above. Check each service's current pricing page, and ask an engineer for a quote on your actual project — album rates and single rates are usually structured differently.

Choosing between them

Automated mastering earns its place when you release often, when you are working to a schedule, when the material is a demo or a work in progress, or when you simply need a competent, consistent result now. An engineer earns theirs when the release is a milestone, when it is an album that has to hold together, when it is destined for vinyl — where the cutting stage has real technical requirements — or when you want someone to tell you the truth about the mix.

Many artists end up doing both, and there is nothing incoherent about that.

Why loudness normalisation changed the stakes

For years, mastering decisions were shaped by the assumption that louder wins. Streaming playback normalisation changed that. Spotify normalises playback level toward a published target of -14 LUFS, and describes on its own help pages how tracks above and below that target are adjusted at playback.[1]

The practical consequence is that crushing a master for competitive volume no longer buys competitive volume. It buys reduced dynamic range at the same perceived level as a more open master. That is not an argument for ignoring loudness — a record still has to feel right next to its peers — but it removes the reason to fight for the last decibel.

We are not stating targets for other platforms here. Normalisation behaviour differs between services and is changed by the services themselves; check the current documentation of whichever platform matters most to you rather than relying on a figure quoted second-hand.

Preparing your files for mastering

This section applies whichever route you take, and it is where more masters are compromised than by any choice between algorithm and engineer.

Export WAV at 24-bit

Mastering processes expect uncompressed audio, and 24-bit WAV is the standard submission format. The extra bit depth gives the process room at the bottom of the range to work without running into the noise floor. If the project was recorded at a higher sample rate, export at that native rate rather than downsampling first — resampling is a decision best made once, at the end.

The reasoning behind this is the same as the wider case for keeping uncompressed audio in the chain: lossy encoding discards information that later processing would otherwise have had available.

Never submit an MP3

This one is worth stating flatly: do not submit MP3 files for mastering. MP3 encoding permanently removes high-frequency content, stereo detail and low-level transient information. EQ and compression applied afterwards do not restore any of it — they amplify what the encoder left behind. The result tends toward harsh, thin, or oddly brittle.

An engineer will usually reject an MP3 submission outright. An automated service will happily accept one, which is worse, because the compromise is invisible until you hear the result.

If an MP3 is genuinely all you have

Sometimes the session is gone and a bounce is all that survives. Converting that file to WAV before submission does not recover anything — the discarded data is not recoverable by any process. What it does is hand the mastering stage the file in the container it expects, which avoids a second decode step and any compensating behaviour a service might apply to lossy input. The better answer, where it is available, is to reopen the session and export a proper WAV from the project.

Headroom and the master bus

• Leave headroom. Aim for peaks a few decibels below full scale. The mastering stage needs somewhere to work.
• Take limiting off the master bus. Unless the mix genuinely depends on bus compression as a creative element, bypass any limiter, maximiser or loudness processor before you export.
• Check for clipping. Even brief excursions past full scale create distortion that later processing will make more audible, not less.
• Send references. Naming two or three commercial tracks you are aiming at gives an engineer — or a reference-matching algorithm — something concrete to work toward.
• Label the files. Artist, title, and a clear "pre-master" marker. Confusing a pre-master with a master is a mistake that reaches the store.

For the underlying format argument in a production context, our WAV versus MP3 guide covers what each format does to the signal and where the difference genuinely stops mattering.

The bottom line

Automated mastering is a real tool, not a toy, and it has removed a genuine barrier for artists releasing on their own. It is fast, repeatable, and good enough for a great deal of released music. What it does not do is exercise judgement, argue with you, or hear a record as a record.

Whichever you choose, the input decides the ceiling. A 24-bit WAV exported with headroom from a mix you are honestly happy with will get more out of either route than any amount of processing applied to a compromised file.