Broadcast Audio Is Broken. Here's How We Fix It.

The state-of-the-art for sonic excellence has advanced dramatically over the past 10 years, driven in large part by the rise of digital streaming services. Unfortunately, too many radio stations are powered by the legacy of past technologies and past decisions, and the result is bad audio playing out over the air. Super Hi-Fi aims to change that, using the latest available technology and a focused determination to ensure a pristine signal chain: from the moment a song is ingested into Super Hi-Fi’s system all the way to when the final signal hits a car’s dashboard.

‍

Here’s how.

‍

A Bit Of History

‍

To start, some history. The team that makes up Super Hi-Fi has been working for 25 years at the bleeding edge of digital music experiences. We’ve leveraged that experience to create an entirely new kind of signal chain that today powers the best sounding radio experiences on Earth. From Sonos Radio HD (the only full-range, uncompressed, ultra high-fidelity radio product in the world) to Peloton and beyond, our clients demand pristine audio at scale.

‍

To be fair, many of these experiences are in the digital domain, however we believe that broadcast radio companies deserve the same high quality experiences as the best digital services offer. Now, you might ask yourself “don’t radio stations already have great sounding audio?” In general… nope. 

‍

Your Catalogs Probably Don’t Sound Good

‍

By now, we’ve worked with dozens of broadcasters on their digital and broadcast stations. The first thing we do with a new customer is to ingest their catalogs into our system. A part of that process is a sophisticated deep-analysis of the catalog to ensure we have an accurate view of the quality of the music we’re ingesting. What we’ve found is pretty shocking.

‍

On average, we’ve determined that approximately 50% of the music we are delivered by radio stations started out as low-quality MP3 files. The tracks we’re sent are tagged as WAV files, and so our customers think they have a clean, high fidelity catalog. But when you peer deeper into the audio itself, most of it is actually MP3, varying in quality from 96k to 320k. These catalogs were built up over the years, merged in through station acquisitions, or delivered by music servicing companies that used bad audio. But the files came in a WAV, not as MP3, so why the disparity? Because old-fashioned playout systems only support WAV files, so someone at some point just saved the shitty MP3 files as WAV files so they could play back. Yet the quality never got any better.

‍

The other 50% of the tracks are on average actual WAV files. However most of those files have had some sort of audio processing applied, most of them are pre-compressed and not the full-range you would want as a starting point. All-in, the average of the catalogs we’ve analyzed have fewer than 20% of the total tracks as actual CD-quality masters with full fidelity. 

‍

So, if 80% of your catalog is either an MP3 file or a pre-compressed audio file, how do you expect that the downstream quality is going to reflect the best you have to offer?

‍

Dynamic Range Actually Matters

‍

Okay, it gets worse. There’s a phenomenon in pro audio called “the loudness wars”. That term refers to how music has been mastered over the past 2 decades to sound louder and louder. The thinking has been that louder sounding music will stand out more, and sell more, and so over the past 20 years mastering engineers have been finding new ways to create a ‘louder’ sounding track.

‍

The reality, however, is that there’s no such thing as louder with mastering. There’s a peak available signal, and that’s the maximum that any moment of audio can hit. So what’s really happened is that mastering engineers just reduced dynamic range - the gap between the quietest parts and the loudest parts of a song - and smashed everything together near the peak of the available volume scale. Yes, it sounds louder. But it’s really just an overcompressed music file with no finesse. 

‍

Where it becomes a problem is when you play some songs from recent times (where audio sounds really loud), and then songs from the 80’s and early 90’s (where songs had much more dynamic range). The songs with more dynamic range will sound much quieter overall, but because their peaks will still be near the top of the available range, there’s no way to turn them up any louder. So the way that radio companies have generally handled this has been to compress those files to reduce their peaks, and then turn them up to more closely match the volume of the newer songs that are louder but which have less dynamic range. Basically they are dumbing-down the higher dynamic range songs to stay more even with the lower dynamic range tracks.

‍

They’re using the newer tracks as the reference, and bending the older tracks to that new low dynamic range standard. The result is that pretty much all the music sounds flat and lifeless. 

‍

What’s Playing Out Over The Air

‍

So here’s what we’ve seen as the output of many broadcast radio stations: tracks that are in many cases low bitrate MP3 files. Then the high dynamic range tracks get pre-compressed so they’re similar in overall level to the modern highly compressed tracks. So you have the sound of a bad codec, and a bunch of dull-energy music. 

‍

Next, the broadcast standard is to set your processor to -12 on the inputs, so that any errant peaks don’t clip. That reduces the headroom even more, and since each -6db reduces bit depth by one, you now have effectively reduced the signal to a 14-bit audio signal. 

‍

Finally, you configure your audio processor to work its magic, using the AGC to keep the level even; compressing and EQing as required to make the signal sound like radio. But garbage in is garbage out, and no matter how effective your processing is, it’s not going to add in the fidelity that was lost way up the signal chain.

‍

The Streaming Services Have The Answer

‍

Unlike radio broadcasters, the big streaming services have built an ecosystem of quality from the ground-up based on the most modern capabilities available. Spotify and Apple Music have spent endlessly deciphering the code for how to deliver music experiences at the highest fidelity and with the most reliable connectivity. To be clear, we know both companies very well, they are extremely focused on the overall quality of their audio, it is the single most essential aspect of their services. And if you look at their success as businesses, their decision-making is clearly working with consumers. So imagine teams of PhDs in audio; digital signal processing; and consumer experience… all working together to create the right ingredients for these digital music services to continue to deliver the best possible listening experiences. And here’s what they’ve come up with.

‍

They all start with label-delivered music catalogs. Pure, unadulterated music delivered with zero audio modification, precisely as it was delivered from the masters. Every single track in their catalog is stored as a 16-bit, 44.1k digital file (and even higher if delivered in high-res audio). That’s the starting point.

‍

Next, the services need to encode the audio to bring down the bitrate for delivery. To these services, “quality” is actually the combination of fidelity + reliability, so it’s essential to reduce the filesize to the lowest possible bandwidth before it impacts the audio fidelity. That way the music will keep playing even on compromised networks. On this front, Apple is the clear leader, offering the widest array of options, including Dolby Atmos, Full Lossless, and High Fidelity Low Bandwidth. For that last option, they selected AAC Low Complexity 256k. 

‍

Why AAC-LC? Because it is the only codec that at medium bitrates (256k to 320k) has been proven beyond any doubt to be effectively indistinguishable from the uncompressed source file. We want to be clear: Apple - and others - have done tremendous research in this field, have dedicated so many resources to this question, and yet all have come to the same conclusion: no ability to perceive differences in the audio fidelity between the encoded and the original source version, no matter the track or the reference system used to test it on. The best available example of this is the set of tests performed by the BBC when they were working to select a codec for their music services. Definitely worth a read.

‍

https://downloads.bbc.co.uk/rd/pubs/whp/whp-pdf-files/WHP384.pdf

‍

Finally, how do Apple and Spotify deal with the differences in dynamic range between older and newer song masters? Especially when users can build playlists of songs that vary all over the place? Well, unlike radio companies, who as we’ve explained compress the older music and turn it up, the streaming services do the exact opposite: they leave the music untouched by any compression, and they turn the newer music down to match the average perceptual volume of the older, more dynamic tracks! The result is that the tracks stay fairly level with one another, song into song, but the tracks with lots of dynamic range still sound incredibly punchy and full range. No flat or dull experiences there - instead it’s full range, full fidelity, and yet still smooth and even sounding from track to track. 

‍

One really important technical point: note above that we referred to the ‘average perceptual volume’. The standard measure for volume in the radio world is RMS, which refers to the average power of a signal over time. But the streaming services actually measure LUFS, which is the perceptual loudness of a signal. It's how humans hear loudness. So when Apple or Spotify adjusts the relative volume of the tracks against each other, they use the LUFS standard to determine how loud a track is, not RMS. Apple calls their volume technology SoundCheck. Spotify used to call theirs ReplayGain, however they now just refer to it as LUFS normalization. Either way, they’re doing the same thing and it works incredibly well.

‍

Now, when you look at the waveforms of the LUFS normalization, it is going to look very different from what you are used to seeing when analyzing peak values (see below graphic). In fact it will look backwards from what it’s actually doing. You will see the older, higher dynamic range tracks filling the available space, while the newer ‘louder’ music will look much quieter by example (because it was turned down), and because that newer music is so compressed from the masters those tracks will look like a rectangle, with no peaks or valleys. Overall it will look like the volumes between tracks are all over the place. But it’s an illusion… because LUFS is assessing the perceptual loudness, these songs will sound very even to one another when you actually listen to the playback. Don’t think like a radio engineer, think instead like a Spotify audio expert, and it’ll all make sense. 

Finally, both Apple and Spotify use a very transparent brick-wall limiter on their outputs to ensure that any errant volume peaks are kept below 0db, just as a safeguard against any audio clipping. These limiters are almost certainly very rarely if ever hit, but they’re inline just as a final protection method.

‍

So that’s the streaming services. It starts with cd-quality catalog; it’s untouched by any digital signal processing (eg. dynamics compression); it’s then encoded into the best possible codec for fidelity + reliability; it is LUFS normalized so the volumes are similar but the dynamic range is uncompromised; and then there’s a brick-wall limiter at the end of the chain as a final safeguard. With literally any option available to these services, after billions of dollars spent on technology and user experience, this is what they came up with. And so then the question becomes, why wouldn’t radio companies use all of this research and decision-making, paid for by Apple and Spotify and proven out with their respective commercial successes, to their own benefit? Or to make it simpler: why wouldn’t radio companies just copy what the streaming services have done?

‍

The Super Hi-Fi Approach To Broadcast Audio

‍

Today, we are basically using the streaming services model for audio delivery. After a ton of trial-and-error, after working with our digital partners, after measuring the results of our efforts, this is what we’ve come up with as a formula for the absolute highest quality radio experiences.

‍

1. We always start with authentic, high fidelity tracks, as-delivered from the labels. Every track in our customer’s account has to start as a fully validated 44.1khz, 16-bit master, and then we store the file with no adulteration whatsoever. No compression, no modification. Just the original source file.
2. We apply our own LUFS volume normalization. We don’t compress, we just analyze the audio, and turn the newer tracks down only at mixing time to ensure a consistent volume with the older, high-dynamic range tracks.
3. We use an incredibly transparent look-ahead brick-wall limiter to catch any stray peaks. We wrote this tool from scratch to ensure that it has basically no impact on the output.
4. We deliver to our playout devices at AAC-LC 320k. This is the codec that the BBC tested with, so it gives us comfort that it’s the best available choice for fidelity. And yet it’s efficient enough to be very reliable even for remote transmitter sites that may not have the most stable bandwidth. 

‍

That’s it. That’s all we do. 

‍

The Results

‍

The first result is that the stream we send down the line sounds incredible. Full dynamic range, super consistent, and zero chance of any clipping. But when the audio hits the processor, it gets even more interesting. Now the input gains on the processor can be set all the way up to the top, since any audio peak management happened back at the source. Run it at -1 if you want, so you can use the full available headroom that the audio processor makes available. Better audio in, better audio through the unit.

‍

Next, you’ll see that the AGC moves less. Since the tracks are already LUFS normalized, the AGC doesn’t have to work as hard to keep the audio in a good range. That means less ‘rise and fall’ over time as you listen, for a smoother and less jarring experience.

‍

Finally, when you’re running full fidelity, and full dynamic range, the magic of the audio processor can be fully harnessed. The wizards at Telos and Orban have spent decades making their processors into the ultimate weapons for FM preparation. Why tie their hands behind their backs with low resolution, flat music files? Give them the full juice, and let them do what they were designed to do. And as audio engineers, let your skills really fly and carve the sonic signature for your station that you actually want, unencumbered by the poor upstream audio that’s been plaguing broadcast radio for years.

‍

Copy This

‍

Okay, so that’s our formula for amazing sounding radio. It’s not rocket science, it’s just a practical solution formed from watching what the streaming services have been doing, and combining it with what we’ve learned from the broadcast market. And here’s our final pitch: copy us. Whether you work with Super Hi-Fi or not, this is the model that we believe delivers the best sounding audio at the highest level of reliability. So take it for yourself.

‍

• Analyze your own catalog, and see what the source material is actually comprised of. Don’t allow a single track into your catalog that isn’t a full-range digital master.
• Don’t compress the files, instead use a LUFS normalization routine to level out the volumes.
• Throw a brick-wall limiter inline so you eliminate the possibility of clipping.
• Then, go reset your processor to a higher input level, and rebalance your settings to take advantage of all that new-found frequency and dynamic range.

‍

It’s a different approach for sure from what radio broadcast has been relying on for years. But the technology has improved to the point where it doesn’t make sense to rely on the old way of doing things anymore. Consumers are more demanding, HD Radio has higher fidelity so the differences are more noticeable, and the opportunity to deliver a better experience is well within reach. Together, we have an opportunity to improve the overall quality of today’s radio experiences, so we can attract more listeners; so we can provide a credible and competitive offer to demanding consumers, and so we can help radio companies to thrive.