When I got my first job in public radio almost nine years ago, I was aware that I brought to the position a rare and valuable credential: a recent memory of what it was like to listen to public radio before working in it.
Alas, this precious resource, like so many others, proved nonrenewable. Within a year, I found myself asking the eternal internal question: “Do listeners actually care about [fill in the blank]?”
We try to answer such questions — by soliciting audience feedback, by its nature anecdotal, or by commissioning research, which is expensive and often poorly executed.
But there are moments of revelation when those limited modes of inquiry produce answers so conclusive and compelling that all attuned pubcasters perceive them subconsciously, like a great disturbance in the Force.
Such was the case recently when Current’s Ben Mook posted an NPR-produced chart, displaying in devastating clarity what we’ve all known: Some public radio content is WAY LOUDER than the rest. (Or, more problematically, way quieter, but that doesn’t render as well in ALL CAPS.)
Do listeners actually care? Anecdotal evidence surfaced when On the Media, which according to the chart boasts one of the burlier waveforms, tweeted a link to Mook’s article.
— Galen Broderick (@galkbro) June 6, 2014
— Robert Ashley (@robertashley) June 6, 2014
— Will Lavender (@willlavender) June 6, 2014
— Ian Tornay (@crash7800) June 6, 2014
@onthemedia I appreciate that ur show is loud enough to hear straight from phone w/o the need for earbuds. Your theme song is a bit loud tho
— LoLo Bonobo (@Neural_Coil) June 6, 2014
If you want research, we’ve got that, too. An NPR Labs study last year found that a sudden change of just 6 decibels within a program stream sent half of a study group of 40 listeners diving for their volume knobs. Bigger shifts annoyed them so much that they said they’d turn off the radio entirely.
It’s as if millions of voices cried out in terror and were suddenly silenced.
As a content creator, I’m actually made more nervous by the half of listeners who don’t fiddle with the radio when the level spikes or drops. Picture it: They’re listening in a car or train, a poorly mixed cut of phone tape comes on, and it sinks below the noise floor into oblivion.
Whoever spent their day getting and cutting that tape may as well have called in sick.
Any good reporter knows that if you want the truth, follow the money.
How do I know that climate change is real? Corporations are betting on it. The fossil fuel industry is investing billions in developing Arctic oil and gas reserves that will become accessible only if polar ice retreats at the rate that scientists expect.
How do I know that inconsistent levels are bad for ratings? Commercial radio stations compress their signals into undifferentiated candy bar-shaped waveforms that are equally loud and equally soft, all the time, always and forever. (Yes, I know my use of “loud” here is technically incorrect. I’ll get to that later, nerds.)
This dynamic range compression (not to be confused with data compression) is electronic or digital processing that automatically boosts quiet sounds and lowers loud sounds. (A related effect called “limiting” rolls off only the peaks and leaves the valleys alone.) Compression can be applied in different ways and to varying degrees, but at its most extreme, compression is used to make everything as loud as it can possibly be.
Here’s an example of commercial-style radio compression applied to perhaps the most famous dynamic contrast in the classical music literature. First we hear it raw, then compressed.
Yes, that sounds nasty. Yes, it robs the music of its essential character. But there’s a reason commercial radio does that. It makes for a very easy listening experience, with no knob-fiddling. A super-compressed station can sink into the unobtrusive background of a dentist’s office like wallpaper, and stay there about as long as the literal wallpaper. And that means ratings, my friends.
While many public radio stations apply some amount of overall compression to their broadcast signals and digital streams, it’s usually pretty subtle by commercial standards, and many barely compress at all.
I suspect public radio’s aversion to compression stems from the fact that most stations started out airing classical music, and many still do, at least in some dayparts. Popular music tends to be somewhat dynamically static, but expressive contrasts of loud and soft are as integral to classical music as pitch, timbre and rhythm.
Compression aesthetically warps classical music, and jazz to a lesser extent, much more than it does any other common type of radio content. Yet, ironically, classical music needs compression more to avoid any number of problems.
Julie Amacher, the silky-voiced program director of American Public Media’s Classical 24 service, knows this all too well. C24 is the 98-pound weakling on NPR’s chart of volume levels, shivering in the shadows of beefcakes like WNYC’s On the Media and Radiolab (insert joke about loud New Yorkers here).
“If we have a piece that has incredibly soft parts, what happens then is the sounds sensors go off, the [silence] alarms,” Amacher told me. “Then we’re getting engineers around the country who are having to run out to their transmitters to turn their transmitter back on the air.”
Amacher and her hosts often find themselves adjusting levels during those hushed moments, driving the faders to the max and hoping that something will come out the other end, even if it’s just the violists shuffling in their seats. You could call that a form of manual compression.
To be clear, Amacher is not aesthetically opposed to automatic levels processing. In fact, she’s counting on stations to do it.
“We don’t put any processing or anything on the music that comes to stations, because we want the stations to be able to create their own sound, if you will,” she said. “I mean, every station engineer has their own way of processing that audio.”
Amacher’s approach is the standard one, and it’s probably necessitated by the politics of network-station relations, though I fear that the faith on which it is based may be undeserved. In fact, I think the “someone downstream will fix it” attitude is the root problem behind public radio’s chronically wonky levels.
A lot of producers don’t fuss too much over their mixes, live or pre-produced, because they assume that somewhere along the radio pipeline a magic box is spitting out a perfectly standardized product. All too often, that is simply not the case.
Like many people who work on this Island of Misfit Toys we call public media, I’m a failed artist. I went to music school, studied classical composition and electronic music, and still spend an occasional weekend busting out electro-art pop or semi-ironic covers.
Serendipitously, my training has been amazingly applicable to my actual career. Mixing a radio story is child’s play compared to mixing a record, so it’s not surprising that the best- and most consistent-sounding radio is made by us musician refugees hiding out in broadcasting until someone ejects us like the imposters we are.
Take Radiolab, the most sonically advanced radio program in the history of the medium. Host/creator Jad Abumrad studied composition at Oberlin, and technical director Dylan Keefe spent 17 years playing bass in the alt-rock band Marcy Playground (yes, he’s one of the “Sex and Candy” guys).
The show’s ultra-professional sound results from the union of two habitual knob-tweakers. “I don’t think it’s an accident. I think we speak the same language,” Keefe told me via ISDN from WNYC’s studios, where he and Abumrad obsessively futz with Radiolab’s levels.
“Frankly, that’s what we spend most of our time working on, is how does Radiolab sound in comparison to itself as it goes from piece to piece,” he said.
This is probably a good place for me to offer a definition of “mixing.” Mixing is the art of making all the bits and pieces of an audio composition (musical, journalistic, whatever) sound right relative to each other.
“Jad is a very, very masterful mixer,” said Keefe (himself no slouch at the board). “[Abumrad] is somebody who really knows how to make music out of people talking.”
And that’s what it is — music. Balancing an actuality against a voice track is like balancing the brass section against the strings, and it takes a musical ear to get it right.
Unfortunately, in the era of digital audio editing, too many people use their eye. Whereas the radio mixologists of yesteryear just sat at a board, moving the faders in response to what they heard over the monitors (and maybe an occasional glance at the meters), digital editing presents us with waveforms on a screen that too many people just grow and shrink until everything looks even.
But the waveforms lie. We know this thanks to the Depression-era research of two scientists at Bell Labs — Harvey Fletcher and Wilden Munson. They gave us the Fletcher-Munson curve, known as the “equal loudness curve” in its updated forms, and it’s the most useful psychoacoustical concept you’ll ever learn about.
When I explain Fletcher-Munson to my students at Mercer University, I play them a low tone, say 100 Hz. The speakers in the classroom rumble almost inaudibly. Then I hit them with a piercing tone at 10,000 Hz, and watch as the upright nappers in the back row leap involuntarily, spilling coffee all over the notes they haven’t been taking. Good times.
“Which tone was louder?” I ask. “The second one, duh!” they reply.
Then I turn the laptop around to reveal two waveforms of identical height. The tones were of equal volume. Each vibration had the same acoustical power, sending the same quantity of air molecules rippling between the speaker cones and each student’s eardrums.
And yet, the second sound was WAY LOUDER. The lesson? Volume ≠ loudness.
Loudness is subjective; it’s the way we perceive the strength of a sound. But human hearing sensitivity is hugely uneven across the audible frequency spectrum. Fletcher and Munson did perceptual tests on people to find out exactly how uneven it is, and they came up with a series of curves, indicating the decibels it would take to make tones from the bottom to the top of our hearing range sound equally loud to an average person.
It turned out they didn’t get it exactly right, so the revised equal-loudness curves in this Wikipedia chart are shown in red.
The moral of this story for radio producers is that looking at a waveform on the screen doesn’t tell you everything about how loud it is. Sounds with radically different frequency content — like your interviewee on the phone vs. your host in the studio — can look the same, while one sounds WAY LOUDER.
Of course, we all listen as we mix, and we think we’re clicking and dragging those fader automation lines in response to what we’re hearing. But in my experience, the brain averages out the conflicting input it gets from two sensory sources — our perception ends up half-wrong.
Further complicating matters is that fact that the Fletcher-Munson effect diminishes as overall volume increases. That’s why it’s technically the Fletcher-Munson curves; there are many of them. The louder you crank the volume on your speakers or headphones, the louder the bass will sound relative to the other frequency ranges — the curve flattens.
To demonstrate that concept for yourself, pull up a song that you usually only listen to while exercising (i.e., blasting it in your earbuds to stimulate your adrenal glands). Try listening to it at normal volume, and it’ll sound like a whole new mix. You’ll hear little parts you never noticed; some details will become clearer while the overall sound will be less full.
So here’s another lesson: Mix at the volume at which you think people are likely to listen. People listen to music at all kinds of volumes, but they tend to listen to talk radio at a moderate one. Sure, you can crank it up to get a real close listen, to make sure there’s no repeating door slam or dog bark to give away your lazy ambi loop, etc. But afterward, bring it down into comfortable listening range and adjust your levels accordingly.
And it’s not just frequency; duration alters the way we perceive sound as well. Without getting into another science lesson that I am hardly qualified to give, the ear is kind of like the eye in this sense.
Get up in the night to relieve yourself, turn on the bathroom light, and your eyes will burn until your pupils adjust. Then tiptoe to the couch and turn on the TV. The volume at which you were listening before bed will sound WAY LOUDER at first, until your auditory system adjusts. Same deal, basically.
The point of all this perceptual stuff is: Sound is complicated, so trust your ears. You’re never going to quantitatively factor in something like Fletcher-Munson.
Actually . . . you might, if Rob Byers gets his way.
Hold that last thought for a minute, and let’s define “mastering.”
Whereas mixing is the art of making the elements of an audio composition sound right relative to each other, mastering is the art of making the mixed piece sound right relative to all the other audio in the universe that may come before or after it in a playlist.
The problem illustrated by that NPR chart we began with — Snap Judgment being WAY LOUDER than This American Life — isn’t so much one of mixing. It’s mostly a mastering problem.
“All these individual shows are fairly consistent on a daily basis, it’s just they’re not consistent with each other,” said Byers, a whip-smart APM engineer who spoke to me from St. Paul.
I think he’s being a little generous — I regularly hear poorly-mixed elements within pieces or programs that make me hit the roof. But Byers, as a member of a cross-organizational working group that’s looking at ways to standardize levels on everything distributed through the Public Radio Satellite System, perhaps feels a need to be diplomatic.
“We know we don’t have a problem with production practices. We have a problem with transmission and delivery practices,” he said.
But good delivery starts with good mastering. To wit, here are two promos, both alike in dignity, pulled straight from ContentDepot on the same day. It’s Radiolab vs. Marketplace, WHO’S GOIN’ DOWN, BROTHER?!
Both promos are beautifully mixed, but Radiolab’s is vastly better-mastered. Yes, I just implied that louder is better, and to an extent, I think it’s true. Especially these days.
“In the changing media universe, you know, I’m actually mixing . . . and, at this point, mastering for the uncontrolled environment,” said Keefe, who mixes and masters Radiolab under Abumrad’s creative direction.
“Meaning, if someone puts [Radiolab] on their phone, their iPad, or whatever,” he said, “I cannot assume what was there before or what’s coming after. I can only assume that it’s more varied than the stuff that’s on a public radio station.”
Back in the day, public radio shows could get away with being quiet, as long as they were about as quiet as all the other shows coming over the satellite. You could assume that your show was going to be heard up against another public radio show.
In the podcasting era, a growing share of your audience doesn’t listen that way anymore.
“I can assume that they were just watching a movie on their phone, or that they were just listening to Fugazi, or they were just listening to a book on tape, a number of different things,” Keefe said.
If you want to know just how differently most public radio is mastered compared to popular music, check out this head-to-head. In one corner, it’s This American Life host/creator Ira Glass. In the other, we have “Chandelier” singer Sia.
A direct switch from Ira to Sia on a playlist would be neither smooth nor rare — this is the listening context of the future (and, to an extent, the present).
Of course, loudness is a pretty fraught topic in popular music. For those who are blissfully unaware, a “loudness war” has been raging for at least two decades now, with each artist trying to make sure their song sounds louder than the next one that plays before or after it.
Here’s the video I show my students to demonstrate this concept:
Like all wars, advances in technology have only increased the carnage. Digital multiband compression software can help you make sure that each region of the frequency spectrum is independently maxed out, all the time. Musical subtlety suffers the worst causalities.
Movies have gotten louder too, but since filmmakers like to leave themselves some room to grow for when the plot gets all ’splody, the average loudness is typically less than that of a pop song. I find that books on tape are, amazingly, even more inconsistent in their loudness than public radio shows. So how does Keefe master for a media environment of such random contrasts?
“I’m trying to play it down the middle,” he told me.
Using subtle bits of compression or limiting, but mostly just fastidiously massaging those fader lines, Keefe creates a sound that is even, full, and pretty loud. He still maxes out at -3 dBFS (3 decibels from 0, the highest possible amplitude), leaving radio stations the standard headroom they’ve come to expect that stops the sound from clipping if the pot is a hair north of 0, and also striking the happiest possible medium between maxed-out music and everything else.
This American Life could take a lesson here: You can still sound all hushed without being so damn quiet! Wouldn’t you rather stack up to Sia more like this?
I didn’t compress Glass into Top 40 territory. I just leveled off the most extreme peaks with a limiter and normalized to -3. Sia still blows him out of the water, but at least he puts up a fight.
Of course, we can’t all have a Dylan Keefe at the knobs, so what we need are standards, right? PRSS already has some, as Byers informed me (to my shock).
“Voice levels should be around -15 dBFS, and nothing should ever go higher than -3,” he said. I think that’s probably a little too quiet relative to the broader audio universe, but hey, it’s something.
However, in analyzing content across the system, the working group that Byers is a part of reached a totally un-shocking conclusion: “About half of the programing is out of spec in one way or another,” he said.
And frankly, even if shows were within spec, there would still be annoying variability, because (among other reasons) decibels are used to measure objective volume, not subjective loudness.
So how do we finally fix the problem, once and for all?
“No one is saying let’s strap a compressor or a limiter across everything that goes through PRSS,” Byers reassured me (and by that I mean, all of you). “I think the direction we’re going to head in is finding a new way to analyze the audio . . . that’s perception-based.”
That new and better way has already been the production standard in the European Broadcasting Union for four years now. And for almost two years in the U.S., it’s been the FCC’s enforcement standard for the CALM Act — the federal law that mandates TV commercials can no longer be so much louder than Matlock reruns that they shock Granny right out of her Barcalounger.
Ladies and gentlemen, I give you . . . [drumroll] . . . Loudness Units. “LU” to his friends.
Basically, LUs are decibels graded on a curve — the Fletcher-Munson curve. An engineer watching an LU meter is going to get a much more accurate picture of how loud a signal is than with the old dBs, because it takes human hearing sensitivity into account.
Those wanting more details should watch this amazingly accessible introductory lecture on the European loudness standard by the Austrian TV engineer Florian Camerer (h/t Byers).
A better loudness standard isn’t going to immediately result in the “audio nirvana” that Camerer envisions, or automatically make your programming smooth and in-spec. At least, not all programming.
“For live programming, it’s always going to be up to the content creator, I think . . . to deliver a consistent product,” Byers said.
Live board ops, meet your new North Star: -24 LUFS. That’s what Byers said is the new optimum average level. Set a steady course, and hold it until you’re over the horizon (i.e. forever).
But for shows that send preproduced content to PRSS, there is the possibility that files could be “loudness normalized” at the point of distribution before they’re beamed out to stations. Software could analyze each file to determine its average LU and automatically gain the whole thing up or down to meet an ideal spec.
This is in contrast to the standard “peak normalization” that most radio producers rely on, which only standardizes the file relative to its highest amplitude spike, and therefore usually doesn’t do too much for you. At least use a limiter, man!
“I can hear hackles being raised right now all around the country, because we’re potentially saying, ‘We’re gonna touch your content,’” Byers said, adding (in the same tone of voice in which one would say, “Sir, put the gun and the baby down!”) that loudness normalization is not, repeat, not to be confused with compression.
But concerned parties need not argue over a hypothetical. Just listen to how it works on the popular (with audiences, not artists) music-streaming service Spotify.
“If you go into Spotify, and you go into your preferences, there’s a little tick box in your preferences where you can turn on or off loudness normalization,” Byers said.
With the normalization function on, compressed-to-the-max Black Eyed Peas songs will come down a bit in loudness, but most other stuff will come up, and everything will come out sounding more or less equal. iTunes has a similar function.
Musicians, like myself, are extremely excited about file-based loudness normalization, as it has the potential to be the carnation-in-the-muzzle of loudness warriors everywhere. I don’t have to be louder than you if we’re all going to be made equal in the end! Welcome to the new Workers’ Paradise, comrade!
Byers stressed to me that all this talk about a public radio loudness standard is only preliminary, at least as it pertains to his working group comprised of APM, NPR and PRSS delegates. “My hope is that we’ll pull in a few other content creators and have a larger discussion about what the impact would be for this across the public radio community,” he said. (Byers welcomes feedback via Twitter and email.)
In all likelihood, Byers said, they’re going to have to come up with different standards for music and talk programming.
In the meantime, several provinces of the APM empire are already using loudness metering and normalization, including on the all the little promos and underwriters they feed. As you can see in this chart that Byers provided, these audio odds and sods produced by lots of different people using lots of different methods used to be, predictably, all over the map. Now, he said, the complaints about sudden, brief explosions or crashes in loudness have ceased.
Of course, remember that loudness normalization is a mastering step. Whether you do it yourself for your local station or the Internet, or let Spotify or PRSS do it for you, you still have to, you know, mix your stuff well (which I hope this article has given you a few pointers on how to do).
Nothing I’ve written here is news to professional engineers, but most of the audio production in public radio isn’t done by professional engineers anymore.
There’s a new generation of content creators coming up. Its members (of which I am one) have done all their production themselves since day one on the job, using increasingly intuitive and powerful digital tools. For them, writing and mixing are part of the same integrated creative process, and that is resulting in tremendously exciting work that makes fuller use of the medium. At times, it’s also revealing their amateurism when it comes to audio engineering.
So, take it from me, the (failed) composer: When in doubt, take your eyes off the screen and listen. Then listen to your mix in a fast-moving car or while mowing the lawn. Then go back and fix everything you now know you did wrong. Repeat.
Adam Ragusea is the outgoing Macon bureau chief for Georgia Public Broadcasting. In August he will become Visiting Professor and Journalist in Residence at Mercer University’s Center for Collaborative Journalism.
Copyright 2014 American University