Why stations process

From the broadcast.audio wiki · Chapter 01 of 9

Part of a series on
Broadcast audio processing

Chapters

Reference

Contents

1The car is the design target
2The receiver is not your monitor
3A playlist is not an album
4Why "just play the records" fails
5The competitive dimension
6What this means for a chain
7See also

A record sounds fine in a control room. The trouble is that it never arrives in a control room.

It arrives in a car doing 90 on a motorway, with tyre roar and wind noise and an engine and a fan, and someone in the back seat talking. It arrives on a kitchen radio next to a running tap. It arrives on a phone speaker the size of a fingernail, in a pocket, in a bag. It arrives at the far edge of a transmitter's coverage where the hiss is louder than the strings. That environment is not a nuisance around the edges of the job — it is the job. Broadcast processing is the set of things you do to a signal so that it survives being heard in the places people actually hear it.

The car is the design target

For decades, the car was the primary listening environment for radio, and it still is for a lot of stations. It is worth being concrete about how bad it is.

The noise floor in a moving car sits somewhere in the region of 60 to 75 dB SPL, mostly low and low-mid, rising with speed. It is broadband, it is constant, and it varies by 15 dB over the course of one journey depending on road surface. Against that, a listener sets the volume by the loudest thing they hear. Every part of the programme quieter than about 15 dB below that point is, for practical purposes, not there. It isn't quiet — it is inaudible.

Now put a record with 20 dB of internal dynamic range into that. The chorus is fine. The verse is gone. The intro is gone. The quiet outro that the producer spent two days on is gone, and it takes the last four seconds of the song with it, which is where your station's name normally goes.

The dynamic range a recording has and the dynamic range that survives to the listener are two different numbers, and in a car the second one is small. Processing is, at bottom, the business of taking the dynamic range that exists and fitting it into the window that survives. Everything else in this wiki is detail on how.

The receiver is not your monitor

The second half of the problem is the equipment. A car head unit feeds small drivers mounted in door cavities that were designed by someone thinking about door hinges. The speakers are rarely a matched pair in a sensible position, the low end is whatever the door resonance gives you, and the "stereo image" is a driver with one ear near a tweeter. Kitchen radios and portable receivers are mono, with a single small speaker that does essentially nothing below 200 Hz. Phone speakers are worse and are used constantly.

None of these reproduce a mix the way the mix was made, and the failures are not subtle: a gentle low shelf is inaudible on half of them, a wide stereo effect partially cancels on anything that sums to mono, and the bottom octave that anchors a track on a proper system does nothing on a phone while still consuming headroom on the way there.

And on FM there is reception on top of that: weak signal, multipath in a city, a receiver blending toward mono to save its signal-to-noise ratio, hiss in the top end where the noise lives. The louder and denser your programme is relative to that noise, the less of it the listener notices. That is not an aesthetic preference; it is an SNR argument, and it is the reason broadcast processing exists at all.

A playlist is not an album

The third problem has nothing to do with the listener's environment and everything to do with the source material.

An album is mastered as a unit by one person who made the levels consistent on purpose. A radio playlist is not. In one hour a station plays a 1968 single transferred from a tape, a 1994 CD master, a 2021 release mastered to be as loud as it could possibly be, a jingle that was deliberately made hotter than everything, a commercial that was definitely made hotter than everything, and a live announcer on a microphone in a room with a fader that someone else set.

The spread in average loudness across that set can easily be 15 dB. The spread in tonal balance is worse and less fixable by turning a knob: the old transfer is dull and narrow, the 90s master is scooped, the modern one is bright and already squashed flat. Put them back to back at matched level and they still don't sound like the same station. They sound like a shuffle of someone's hard drive.

This, not the car, is the reason I run processing on an all-digital web station where there is no transmitter, no terrain and no hiss. The playlist problem does not go away when the delivery medium becomes perfect. It is the one part of the job that has nothing to do with FM.

Why "just play the records" fails

The argument against processing is simple and, on its own terms, correct: the record already sounds the way someone wanted it to sound, so leave it alone.

What that argument leaves out is that the listener does not experience the record. They experience the sequence. And in the sequence, "leave it alone" produces a stream where the listener has to work: turn it up for the old song, turn it down for the ad break, turn it up again when the announcer comes in too quiet. People do not do this. They do it twice and then they do something else — and in radio, "something else" has historically meant the next button on the dial.

That is the whole commercial case and it is worth stating without dressing it up. A listener who reaches for the volume control is a listener whose hand is already on the radio.

The competitive dimension

There is a second, less respectable reason, and everyone in broadcast knows it: on a dial full of stations, the louder one sounds better.

This is a well-documented perceptual effect and it is unfair. In an uncontrolled comparison the louder of two signals is consistently judged clearer, fuller and more present, by trained and untrained listeners alike, up to fairly extreme amounts of processing. A listener scanning the band is performing exactly that comparison, several times a second, with no level matching — so a station 2 dB louder than its neighbour wins before anyone has heard a note.

So stations got louder, and because the only way to get louder without exceeding the transmitter's hard modulation limit is to reduce the ratio between peak and average, they got denser. This has been going on considerably longer than the loudness war in recorded music, and with far less public argument, because almost nobody outside the industry knew it was happening.

It has a cost, and it is worth being blunt about what the cost is. Past a certain point a station is no longer playing records. It is re-mastering them on the way out the door, in real time, to a house tonal balance, and returning them to the listener noticeably different from how they were made. Independent artists who hear their own track come back off a heavily processed station usually notice immediately. Some of them love it. Some of them are furious. Both reactions are reasonable.

What this means for a chain

Everything that follows in this wiki is an answer to one of the problems above, roughly in this order of importance:

Consistency of level over time, so nothing disappears and nothing jumps out. This is levelling, and it is 80 % of the perceived benefit. If you only do one thing, do this one. (Chapter 2.)
Consistency across the spectrum, so a bass note doesn't duck a vocal and a dull master doesn't sound dull relative to a bright one. (Chapter 3.)
Peak control, so that the loudness you gained isn't given straight back to the one transient in the track. (Chapter 5.)
Stability of the stereo image and the low end, so the result behaves the same on a phone, a car and a pair of headphones. (Chapters 6 and 7.)
Character, which is the part everybody argues about, and which on FM was never separable from the engineering anyway. (Chapters 4 and 8.)

TRANSMIT Tower's preset list runs from a restrained setting that barely does anything beyond levelling to settings that are frankly aggressive, and that spread is deliberate. The trade-off described in this chapter is real, both ends of it are defensible, and which end is right depends entirely on what you are making. A thirty-second promo is not a classical recording.

Knowing why each stage exists is what lets you decide how much of it you need. That is the point of the rest of this wiki.