Finishing the Mix: Space, Automation & Translation

Every mix reaches a moment where nothing is obviously wrong and it still isn't done. The drums hit, the vocal is clear, the bass is solid. But the chorus doesn't lift, the voice sits on the speakers instead of in the song, the guitars vanish on a phone, and when you press mono a whole instrument disappears. The last twenty per cent of a mix is where these things get fixed, and none of them is a new instrument or a new plug-in. They are questions about space, movement and checking.

This is the second half of the Mixing & Mastering specialization and the piece that closes the mix. It covers depth and width, what to do on the mix bus, automation, the checks that tell you whether a mix will survive the world, how to handle revisions, and how to print and deliver the finished mix. It ends with the longest checklist of the course: twenty things to confirm before you call a mix done.

Depth: how to put things in front of and behind each other

A stereo mix has two obvious dimensions, left to right and loud to quiet. The third, front to back, is made by the same cues the ear uses on real sources. A sound that is close is louder, brighter (air absorbs high frequencies over distance), mostly dry (the direct sound outweighs the room's reflections) and arrives well before its reverb. A sound that is far is quieter, duller, wetter and its reverb arrives almost with it. An engineer therefore moves a sound back with four tools: a little less level, a gentle high-frequency roll-off, more reverb or early reflections in the blend, and a shorter pre-delay. To bring it forward, they do the opposite: less reverb, a longer pre-delay so the dry sound stands alone for a moment, and a touch of presence. The 6 dB per doubling of distance that physics gives for a point source in open air (Microphones & Recording, 36/61) is far more than a mix needs; in practice a few dB, a few kilohertz and a different send level are enough.

Depth is also a budget. If every sound is far back, nothing is forward; the lead voice and one or two other elements sit at the front and the rest are arranged behind them. Early reflections, the first few echoes from a room's surfaces, tell the ear about a space's size and are shorter than a full reverb; a small-room reflection set can place a sound in a space without a long tail that would blur the rhythm. Try the model below. It is a set of typical moves, not a simulation of air.

Width: the stereo field and the mono test

Width comes from putting different things on the two sides. Hard-panned doubles of a guitar part, a stereo keyboard, overheads and room microphones, a short reverb or a modulation effect each add it. But width has a price, and the price is paid in mono. When the left and right channels are summed, anything that is identical on both sides adds to a louder, centred sound, and anything that is opposite cancels. A phone speaker, a smart speaker, a club's single-point system, a TV's mono dialogue path and a radio in a car park are all, for practical purposes, mono. A wide mix must survive the fold.

The meter that shows this is a correlation meter. It compares the left and right signals sample by sample and reports a number from +1 (identical, so mono) through 0 (unrelated) to −1 (opposite polarity, so they cancel in mono). A healthy music mix lives mostly between 0 and +1, with the low end firmly near +1 because bass is usually in the centre. A reading that stays negative for long is a warning, not a verdict: check by switching to mono. Another view splits the signal into mid and side: mid = (left + right) ÷ 2 is what the speakers share, and side = (left − right) ÷ 2 is what differs. Mid/side EQ and level controls let you widen only the sides, or high-pass the sides below about 100 to 150 Hz to keep the low end centred (Mixing Bass & Guitars, 36/86).

The classic trap is the Haas-delay widener: copy a mono guitar, delay one side by 10 to 20 ms and pan the two apart. It sounds wide and it folds to mono as a comb filter, with notches at odd multiples of 1 ÷ (2 × delay). With a 12 ms delay the first notches are at 42, 125 and 208 Hz, right in the guitar's body. A true double is a second performance, so its differences are musical and random, and it folds to mono without a comb. The tool below uses both, plus a polarity error.

Two practical rules come from this. Keep the sounds that carry the song (kick, bass, lead vocal, snare) in the centre, where they are the same in both channels. And check mono often, not once at the end, because a mix that has been built wide can be hard to rescue.

The mix bus: process it, and know when to commit

Everything is summed at the mix bus, so a little processing there changes everything together. The usual moves are a gentle bus compressor to glue the mix, a small broad EQ tilt and sometimes a saturator or tape model. Published guides agree that this is a gentle job: a low ratio (about 1.5:1 to 4:1) and a few decibels of gain reduction at most, with some of them recommending a slow attack so the transients get through and some arguing for a fast one. They disagree on attack, so settle it by ear. Some engineers mix into the bus chain from the start (the top-down order in Session Prep, 36/84); others add it near the end. Either way, put it on early enough that you balance against it, because adding a compressor to a finished balance changes the balance.

The delivery question is whether the bus processing goes on the file you send for mastering. Mastering engineers generally ask for a mix with no limiter on the bus, because a limiter permanently removes peaks they might want to shape, and many want the glue compressor left on if it is part of the sound. The safe habit is to bypass the limiter, and print two versions if you are unsure: one with the bus chain and one without. To commit is to decide that a choice is final: to print the reverb into the track, to print the processed vocal, to bounce the guitars. Commit when the choice is made, to stop second-guessing and to protect the sound from a plug-in you may not have next year. Save the uncommitted version first.

Automation: the mix that moves

A static balance is right for one moment in a song. A song has many moments. The verse needs the voice intimate and the chorus needs it to lift; a guitar fill should pop out for two bars and then go back; a snare reverb might throw on the last beat of a phrase. Automation is the recorded movement of a control over time: level, pan, an effect's send, a filter. The fader rides of a mixer's hand are now a lane in the DAW.

Work in passes, in an order. First do section moves (the verse level versus the chorus level of each group), then vocal rides to even the words that duck behind the band, then mutes to clean the ends of phrases, then effects throws (a delay on one word, a reverb swell into a chorus). Rides done before compression are different from rides after: a ride before the compressor changes what the compressor hears, which is why vocal engineers ride before compressing (Mixing Vocals, 36/87). In DAWs the vocabulary is similar: REAPER, Pro Tools and Logic offer Read, Touch, Latch and Write modes, which differ in whether a fader writes only while you hold it, until you stop, or all the time; Ableton Live uses automation lanes drawn or recorded from controls; FL Studio uses automation clips.

Listen for the voice relative to the band in each half. A flat lane leaves the vocal swamped in the louder chorus; a good ride keeps the difference between voice and band roughly steady, which the listener hears as "the voice is always clear". Automation should be nearly inaudible. If you can hear it work, it is too much, unless the move is the effect.

Translation: checking the mix against the world

Your studio speakers are the one place the mix will not be played. Mixing & Signal Processing (36/63) introduced translation: switch to the places listeners are. The list is short and the ritual matters more than the order. Check mono. Check a phone speaker, where the bass below a few hundred hertz is simply not made, so a bass line is heard only through its harmonics and the vocal level becomes obvious. Check earbuds and headphones, where the stereo image is exaggerated and a sharp resonance can hurt. Check a laptop, a car (where the cabin boosts the low end and road noise buries the quietest parts), and at a low level, as in a kitchen, where the balance should survive because the bass and the treble fade first. The Yamaha NS-10 story is this idea.

The filters below are approximations of typical devices, built from standard biquad filters so you can compute what survives. Real products vary widely: use them as a thinking tool, then check on actual devices. The meters compare the synthesised mix before and after each filter.

If the phone check loses the bass, the fix is not more bass. It is harmonics: a little saturation or a mid-range sibling of the bass note, so that the ear reconstructs the pitch (Mixing Bass & Guitars, 36/86). If the car check is boomy, the bass is probably too strong between 60 and 100 Hz in the mix. Make small moves and recheck rather than swing the whole balance.

Fresh ears, revisions and client notes

After hours with a mix the ears normalise everything, and what you hear at hour six is not what is there. Take a break of at least a few minutes every hour and, for the final decisions, leave the mix overnight. When you return, listen once without touching anything and write your list before you open a plug-in.

A revision round is a conversation between two people who hear differently. If you are giving notes, be specific and place them in time ("the second chorus: the guitar covers the last word"), describe the problem rather than the knob ("I can't hear the words" rather than "turn the vocal up 2 dB", since the fix may be elsewhere), say what you like so it is not lost, and send one consolidated list. Listen on the same kind of speaker as the engineer's reference if you can, and say what you used. If you are receiving notes, ask for a reference track, ask which version they heard, and recall the session rather than start over. Save each revision as a numbered version and keep a log of what changed. Agree the number of revision rounds before the project starts.

Printing the mix: stems, alternates and delivery

A finished mix is exported as a file. Export the whole song from its first bar to the end of its tail, at the session's sample rate and bit depth (24-bit is the usual, or 32-bit float), with no dither if you are not reducing the bit depth, no normalisation and no lossy format. Leave headroom: guides for sending mixes to mastering typically suggest peaks around −6 to −3 dBFS, and the mastering engineer will bring the level up. Do not leave a limiter on the mix bus unless the master is intended to sound like that, or deliver a second version with it, labelled. Name the file so it explains itself: artist, song, mix version and format, such as Artist_Song_Mix03_24-48.wav.

Beyond the main mix, deliveries often include alternates. An instrumental has no vocals. A TV mix is the full mix without the lead vocal, kept with the backing vocals, so a presenter or a dubbing editor can use it. An a cappella is vocals alone, with their effects. A clean mix has explicit words removed or replaced. Stems are group exports (drums, bass, music, vocals, effects) that each start at the same point, run to the same length and sum back to the mix; they let a mastering engineer or an editor adjust a family without recalling the session. Print each one from the same session, with the bus chain on or off consistently, and check that they sum to the original by flipping the polarity of one and watching for silence.

Starting points, not rules

Your ears overrule these. A mix that needs 6 dB of bus compression to feel right is telling you something about the balance underneath; and the most common reason to break the headroom rule is a mastering engineer who asks you to.

Common mistakes

Genre and world differences

In rock and metal the width is in hard-panned guitars and drum overheads, and the mono check mostly asks whether the guitars survive; the low-mid area between 200 and 500 Hz fills quickly. Pop mixes are often very wide in the chorus and narrow in the verse, which is automation of width and level at section boundaries. In hip-hop and trap and in reggaetón and dembow the vocal sits at the front and mostly dry, the 808 or kick dominates the bottom, and the translation check is what the 808 does on a phone, where only its harmonics are heard. Cumbia and other percussion-rich Latin American styles spread many percussion lines across the field around a firm centre; Afrobeats and amapiano productions are likewise built from layered drums and percussion with a vocal on top and a bass that must translate. Jazz, acoustic and folk records usually get depth from the room itself and want very little added; mono compatibility matters less because the stereo picture comes from real microphone pairs, though that deserves a check. Where phones and small portable speakers are the main way people listen, as they are in many markets, the phone check is the most important one. Where music is heard on large club or sound-system rigs, as in Jamaican sound-system culture and many dance scenes, the sub-bass translation matters just as much, so check on a big system if you can.

The final checklist: before you call it done

The longest list of the course: twenty confirmations, in the order you do them. Tick as you go, copy it as text, or print it.

Practise on a real session

Take any multitrack from the free Mixing Secrets library at Cambridge-MT (described in Session Prep, 36/84; use its search page to find a style you know). Mix it, then run only this piece's work on top: check mono and the correlation, push one element back, write one vocal ride, run the translation list, and print the main mix with a stem set. The free library is meant for mixing practice; read its terms before you post a result.

What a graduate just sees

• Depth is made with level, brightness, wetness and pre-delay together; not everything can be at the front.
• Width is paid for in mono: identical content adds, opposite content cancels, and a delayed copy folds into a comb.
• Correlation runs from +1 to −1, and the centre carries the song.
• Mix-bus processing is gentle and in from early on; whether to commit is a decision, and the original is saved first.
• Automation is the mix moving: sections first, then rides, then throws, and it should barely be heard.
• Translation is a ritual of mono, phone, earbuds, laptop, car and low level, and the usual fix is harmonics, not more bass.
• Revision notes describe symptoms and are time-stamped; a break is a tool.
• Deliver at the session's rate and bit depth, with headroom, no limiter by habit, and stems that sum.

Where it shows up next

• Mastering & Loudness (36/64) and Mastering in Practice (36/89): the file you print is what a mastering engineer receives, and the headroom decision here is theirs to use.
• Sound for Film, TV & Games (36/67): stems and TV mixes are standard deliverables there, and mono compatibility matters for broadcast.
• Live Sound Reinforcement (36/66): the translation problem of one mix across many rooms and systems.
• Producing Records & the Music Business (36/70): revision rounds, client notes and delivery are part of the producer's job.
• In the world: the same mono and loudness checks apply to a podcast, a video edit or a voicemail greeting.

If you want to go deeper

• 📘 Textbook. Roey Izhaki, Mixing Audio (Routledge, 3rd edition), on depth, width and automation in detail; Mike Senior, Mixing Secrets for the Small Studio (Routledge, 2nd edition, 2018), on finishing and translation for home studios.
• 📘 Short and practical. Bobby Owsinski, The Mixing Engineer's Handbook (Cengage), on the elements of a mix and delivery.
• 🌐 Free practice. The Cambridge-MT free multitrack library, to run the checklist on a real session.
• 🌐 Free reading. Bob Katz, Level Practices, on monitoring and levels.