A band finishes a mix after three months, plays it in the car, and hears three things at once: the bass is a bit loud, the second song is quieter than the first, and nothing else they know sounds this soft on the radio. All three are real. Two of them are mastering questions, and the third is a trap: "louder than the radio" is how the loudness war began, and the piece before this one (36/64) explains how streaming ended its reward.
That earlier piece covers what LUFS and true peak mean and what each service does with them. This one assumes you know and goes to the desk: how to receive a mix, what a sensible chain looks like, how to treat an album as one object, and how to deliver files that pass every checkpoint between you and a listener.
- Who takes it
- Part 2 of the Mixing & Mastering specialization, inside the Music Production specialization of Music; also for Sound Engineering students. Bands mastering their own records, and anyone who will brief a mastering engineer, need it.
- What it is for
- Making a finished mix sound consistent and ready on every system, and delivering it in the right files, with the right data, for each place it will be heard.
- Comes after
- Mastering & Loudness (36/64) and Finishing the Mix: Space, Automation & Translation (36/88).
- Leads to
- The end of the specialization. For getting the finished record out and paid for, see Producing Records & the Music Business (36/70).
- What this piece gives you
- A procedure, a loudness and limiting demo, an album tool, a vinyl side-length check, and a checklist. It will not make you a mastering engineer: that takes years of listening in a good room. Easier first read: Music Technology & Production (35/73).
Receiving the mix: what to ask for, what to check
The first mastering decision happens before any processing: whether the mix is ready. Mastering engineers generally ask for the mixes as uncompressed files (24-bit WAV is the usual), at the session's own sample rate, with no limiter on the mix bus (or one version with it and one without), and with a little headroom: a few decibels below full scale is a common request, and figures such as −6 to −3 dBFS circulate, though each engineer says what they prefer. Ask them.
Then check what arrives. Are any files clipped? Do all the songs have the same sample rate and length of lead-in? Is there a stray fade, a click at the end, a mono file that should be stereo? Any of these is cheaper to fix at the mix than at the master. If something is wrong with the mix itself (a harsh vocal, a boomy bass) the right move is usually to send a note back to the mixer, as covered in 36/88, not to patch it with a mastering EQ.
Listening first, and deciding what not to do
Listen to each song start to finish before touching anything, on the best speakers you have, at a moderate level, and write down what you hear: tonal balance (too bright, too boomy, thin), dynamics (squashed, or ragged), stereo (narrow, or hollow in mono), and flaws (clicks, noise, abrupt ends). Then decide for each note whether it is a mastering job or a mixing job. The most useful skill in mastering is the list of things you decide not to do.
One rule runs through everything: gain-match every comparison. A louder version sounds better for a moment, whatever it is doing to the music, so a "before and after" that differs in level is not a comparison. Mike Thornton's column in Sound On Sound (2018) says it plainly: set the gain so each version plays at the same perceived level, and "if your ears disagree with the meters, trust your ears".
The chain, and the limiter pushed harder
A typical chain is short, and each stage is optional. Corrective EQ first: Thornton suggests keeping moves subtle, with low Q and no more than about 3 dB of cut or boost, checking the muddy region around 150 to 250 Hz, and treating a lift near 5 kHz for vocal clarity or near 15 kHz for air as examples to try, not rules. A gentle compressor next, only if the mix needs gluing. Saturation or mid/side processing (36/88 explains what mid/side does) only where you can name the problem. The limiter goes last: Thornton's rule is that it should be the final plug-in in the chain, and it should be one whose meter shows inter-sample peaks, because lossy codecs can distort even when no single sample touches full scale (36/64).
The tempting step is to turn the limiter up. Try it. The invented eight-second mix below has a kick, snare, hats, bass and a chord, and a brick-wall limiter with a 5 ms look-ahead and a ceiling of −1 dBFS. "Drive" is how many decibels you push into it. The tool reports loudness on a BS.1770-style scale, then lets you hear the pushed version both as it comes out and turned down to match the original.
Sound isn't available in this reader; the numbers still work.
The loudness figures are a BS.1770-style approximation, not a certified meter: K-weighting with the published 48 kHz filter coefficients, 400 ms blocks with 75% overlap, the two gates, and the mono signal treated as identical in both channels. (A 997 Hz sine at −20 dBFS reads −20.0.) "Peak to loudness" is the highest sample minus the integrated loudness. The limiter takes the smallest gain needed over the next 5 ms, releases at the time shown and smooths the result so the ceiling is never exceeded.
Without JavaScript: the original measures −19.5 LUFS with a highest sample of −3.0 dBFS (peak to loudness 16.5 dB). At a drive of 6 dB it reads −15.0 LUFS with up to 4.0 dB of limiting, and matched to the original it is 4.5 dB down, with its highest sample at −5.5 dBFS. At 12 dB the figures are −11.7 LUFS, 10.0 dB of limiting for 31% of the time, and a peak to loudness of 10.7 dB.
Read the matched column. Once the level is the same, what the limiter bought is not loudness but a lower highest peak: the drums have been flattened against the ceiling. Push further and the kick loses its front. Thornton's advice follows directly: do not over-compress to chase a streaming target, because the service will turn the track down anyway, and a more dynamic master that reaches the target usually sounds better.
Album work: order, gaps and loudness across tracks
An album is heard as a run, so the unit of mastering is the whole run. Three decisions matter: the order, the gaps (silence or overlap between songs; the length is a feel choice, so listen to each transition with real pre-roll, which Thornton sets at about 20 seconds), and the level of each song against its neighbours.
On that last one, "equal" does not mean the same number. A ballad mastered to the same integrated loudness as the rocker before it stops being a ballad. Measure each track's integrated loudness (Thornton treats 1 LU as 1 dB) and use the numbers to find jumps you cannot explain by the music, then decide by ear. Try it: reorder the tracks, change the gap, and watch where the steps are. The limit slider is your own tolerance, not a standard.
Eight invented tracks with made-up lengths and integrated loudness values between −9.1 and −13.6 LUFS. "Step" is the difference in integrated loudness between neighbours. "Forced equal" sets every track to the album mean. The CD figure is the original capacity of the format.
Without JavaScript: in the order shown, with 3 second gaps, the album runs 38:45; three steps are larger than 2 LU and the biggest is 4.5 LU, from "Single" to "Ballad". Forcing every track to the mean of −10.7 LUFS lifts the ballad by 2.9 dB.
Formats and deliverables
| Destination | What to deliver and why |
|---|---|
| Streaming and download | The 24-bit WAV master at the session's sample rate; a true-peak ceiling of −1 dBTP, or −2 dBTP if the master is louder than −14 LUFS, per Spotify's published guidance (checked 8 November 2026), so the lossy encoder has room |
| CD | 16-bit, 44.1 kHz stereo (the Red Book format); dither when reducing the bit depth. A CD was first specified for up to 74 minutes, later discs for 80, with up to 99 tracks. Plants usually ask for a DDP image, a format from Doug Carson & Associates that carries the audio image plus the track and subcode information; a CD-R is the cheaper alternative |
| Vinyl | A premaster for each side, in order, with the sides balanced; see the next section. Ask the cutting engineer what they want before you process for vinyl |
| Immersive | An object-based mix delivered as the service specifies; see the Atmos note in 36/64 |
The data matters as much as the audio. Each recording gets its own ISRC, a 12-character code (two letters for the country, three for the registrant, two digits for the year and five for the designation) defined by ISO 3901 and issued by a national agency or its managers. It identifies the recording, not the song; a remaster is usually given a new one. The release itself gets a barcode number, and CD-Text can carry titles on the disc. Keep a spreadsheet of titles, writers, ISRCs and run times before the mastering session, not after.
Vinyl: what the groove can carry
A record is cut as a groove whose width follows the loudness and whose wiggle follows the bass. More minutes per side means a tighter groove, so a long side must be cut at a lower level. Scott Hull of Masterdisk (2023) puts it this way: "Average cutting studios will tell you: don't go over 20 mins", and on a side longer than 22 minutes the level will need to come down compared with an 18-minute side. He also says he cuts 24 to 25 minute jazz sides every day, that a quiet passage can let a long side be cut louder, and that heavy bass takes more disc space. His conclusion is that there is no fixed rule: the cutting engineer test-cuts the longest side.
Common guidance for the premaster is to keep the lowest bass centred and to watch sibilance, but Hull advises against pre-processing for vinyl (such as rolling off the bass) unless you are certain it is needed. The tool below uses the order from the album tool above and shows where each side lands against the 20 and 22 minute marks. It is a rule-of-thumb check from one engineer's published advice, not a specification.
Uses the same eight invented tracks and the order you set above, without gaps. Two of the tracks are marked bass-heavy. Thresholds of 20, 22 and 25 minutes follow the Masterdisk Q&A cited in the sources; the level cost of a long side is not given as a number there, so none is shown.
Without JavaScript: in the order shown, splitting after four tracks gives side A 18:05, inside 20 minutes, and side B 20:19, just past it, with one bass-heavy track on each side.
Do it yourself, hire an engineer, or use an automated service
All three are used, and none is wrong. Mastering your own record is cheap and teaches you a great deal, but you have heard the mix a hundred times, and your room is the same room you mixed in. A mastering engineer brings a different room, a long experience of what goes wrong, and a second opinion on the mix; it costs money and a few days. Automated services analyse the audio and apply processing; they are quick and cheap, and they cannot ask you a question. A common middle path is to master it yourself, then get another person's opinion (or a service's output) as a reference to compare against, gain-matched, with your own ears deciding.
Starting points, not rules
These are places to start, each tagged with where it comes from. Your ears overrule every one.
| Move | Starting range | Where it comes from |
|---|---|---|
| Headroom on the incoming mix | a few dB below full scale; ask the engineer | Common practice; figures vary by engineer |
| Corrective EQ | low Q, within about 3 dB | Thornton (SOS, 2018) |
| Limiter position | last in the chain, with an inter-sample-peak meter | Thornton (SOS, 2018) |
| True-peak ceiling | −1 dBTP; −2 dBTP above −14 LUFS | Spotify for Artists, loudness page |
| Streaming loudness | about −14 LUFS is Spotify's normalisation level; a quieter or louder master is simply adjusted | Spotify for Artists; see 36/64 |
| Gaps and transitions | judge with about 20 seconds of pre-roll | Thornton (SOS, 2018) |
| Vinyl side | about 20 minutes or less; over 22 the level comes down | Hull, Masterdisk (2023) |
Common mistakes
Pushing the limiter for loudness. The matched comparison shows the gain is mostly a lower peak. Fix: stop where the drums start to lose their front.
Comparing at different levels. The louder one wins. Fix: gain-match, every time.
Processing to justify the session. Fix: write down what you will not touch, and let a master be a small EQ and a limiter if that is what the mix needs.
Setting every track to the same loudness. The ballad turns into a rocker. Fix: use the numbers to find unexplained jumps, then decide by ear.
A ceiling at 0 dBFS. A lossy encoder can distort the peaks. Fix: −1 dBTP, or −2 on loud masters, and listen through the codec.
Mastering from a limited mix. A limiter on the mix bus leaves nothing to work with. Fix: ask for the mix without it.
Leaving the metadata for last. Fix: collect titles, ISRCs and run times before delivery day.
Genre and world differences
What "finished" sounds like differs by style. These are common practices, not measured averages.
- Pop, hip-hop, reggaetón and metal: often mastered loud and dense. With streaming normalisation the extra loudness is turned back down, which is the argument for stopping earlier (36/64).
- Rock, indie and Latin pop: usually in the middle: punchy drums, with a sense of difference between verse and chorus.
- Jazz, folk, classical and acoustic: light touch; the dynamics are the music, and vinyl sides in these styles are often longer because the music is quieter.
- Cumbia, salsa and other dance records: made for dancing; a firm low end and a steady level from track to track matter.
- Afrobeats and amapiano: big, clear low end; check on small speakers and phones, where much of the listening happens.
The mastering and delivery checklist
Mastering and delivery checklist (36/89)
Ticks are not saved when you leave the page; copy the list if you want to keep your progress. To print, use your browser's print command: only the checklist is printed.
Practise on a real session
Cambridge-MT's Mixing Secrets free multitrack library, run by Mike Senior, offers recorded songs split into their individual tracks for mixing practice and teaching, free for educational use; read the terms on its FAQ page before sharing a mix, and contact the rights holders directly for commercial use. Mastering practice starts with a mix, so mix one first (36/87 names three vocal-led sessions: Georgia Wonder's "Siren", Lyndsey Ollard's "Catching Up" and Amber Skye's "Howlin'"), then master your own result with the checklist. Do it twice: once chasing loudness and once matching levels and stopping early, and compare them gain-matched a week later.
What a graduate just sees
- Mastering starts with the list of what not to do; most mixes need little.
- Every comparison is gain-matched, or it measures loudness and nothing else.
- Limiting buys peaks, not loudness; the matched comparison shows the cost.
- An album is one object: tracks are matched to their neighbours by ear, not forced to one number.
- Headroom for the encoder: −1 dBTP, or −2 on loud masters.
- Data ships with the audio: an ISRC for every recording, and the barcode, titles and run times.
- A vinyl side has a length budget, and the cutting engineer decides.
Where it shows up next
- Producing Records & the Music Business (36/70): the metadata, ISRCs and distribution that follow the master.
- Sound for Film, TV & Games (36/67): broadcast loudness specs use the same measurement.
- Live Sound Reinforcement (36/66): gain-matched comparison and a limiter are tools of the PA too.
- In the world: playlists, vinyl pressings, podcasts and the "sound check" setting on your phone all depend on these decisions.
If you want to go deeper
- 📘 Textbook. Bob Katz, Mastering Audio: The Art and the Science (Focal Press), the standard long treatment; and Bobby Owsinski, The Mastering Engineer's Handbook, shorter and practical.
- 🌐 Free reading. Mike Thornton, More Pro Tools Mastering Tips (Sound On Sound, 2018); Scott Hull, How to balance the side times for vinyl (Masterdisk, 2023).
- 🌐 Loudness. Spotify for Artists, Loudness normalization; EBU, loudness pages.
- 🌐 Free course. MIT OpenCourseWare 21M.380 Music and Technology: Recording Techniques and Audio Production (Fall 2016).
Mastering is a short, careful pass: receive the mix, listen, decide what to leave alone, make small moves at matched level, finish with a limiter that protects the peaks, make the album hang together and deliver files and data that pass every checkpoint. If you remember one thing, make it gain-matching.
Try this: take any track, make a copy 3 dB louder, and ask a friend which sounds better without telling them which is which. Then turn the louder copy down and ask again.
Sources. Mike Thornton, "More Pro Tools Mastering Tips" (Sound On Sound, March 2018); Scott Hull, "Q&A: How to Balance the side times for Vinyl?" (Masterdisk, 21 July 2023); Spotify for Artists, "Loudness normalization" (−14 LUFS; below −1 dB TP; below −2 dB TP when louder than −14 LUFS; checked 8 November 2026); Wikipedia, "Compact Disc Digital Audio" and "Red Book (audio CD standard)" (44.1 kHz, 16-bit, up to 74 minutes originally, 80-minute discs, up to 99 tracks), "Disc Description Protocol" (Doug Carson & Associates, proprietary) and "International Standard Recording Code" (ISO 3901, 12 characters, per recording); ITU-R BS.1770 K-weighting coefficients at 48 kHz as used in 36/64. Every number in the interactives was computed in node.
- 📘 Katz, Mastering Audio
- 🌐 Sound On Sound: More Pro Tools Mastering Tips
- 🌐 Masterdisk: balancing vinyl side times