New DSP option idea

[ForSerious]

I've been using the EBU R128 replay gain calculation on all my music for years. For eighty percent of songs, it's good, but the remainder tend to sounds too loud.
On the publicly available website https://www.audiocheck.net/soundtests_headphones.php, the third test down is called the Perceptual Sweep Spectral Flatness Test.
Would it be possible to make another replay gain calculation option based off of the frequency levels from that test?

I admit that I don't know enough about how the EBU R128 is calculated, to know if that test is based off of it already. If that's the case, well there's no point.

[Spoon]

https://en.wikipedia.org/wiki/EBU_R_128

It is likely the gating methods are not working on those tracks, just a guess though.

[ForSerious]

Specifically, I use the Volume Normalize DSP set to EBU R128 [track gain].
It's very consistent in what songs it will make loud: Songs with no bass and no highs (Good example: an acapella solo). I'm guessing any song without the typical loud frequencies under 200Hz.
It also increases the overall volume if the song is half quiet and half loud. The loud half is louder than a comparable all loud song. I'm not saying that that's bad. That's just the behavior.

I have been manually tagging songs that are too loud or too quiet with the amount they need to be adjusted by. My script reads the tags and applies the adjustment accordingly. It's a very manual process to find and tag such songs though. I'm hoping there's a better gain formula out there somewhere.

[schmidj]

Good luck, but you should understand the history of the EBU (and equivalent ATSC) loudness measurements, why they were created and for what purpose, and the (serious) limitations in the measurement algorithms and in perceived loudness in general.

First, my background is that I'm a retired systems engineer, having designed audio systems for a major NY TV network for 36 years before retiring 11 years ago. I was tasked with installing the loudness metering and limiters at our network after the Congress and the FCC implemented the "Calm Act", primarily to "control" "loud" commercials.

Attempts at measuring (and sometimes controlling) loudness go back many years before the current rules and measurements went into effect. There have been conflicting goals here since the early days of recorded audio. Psycho-acoustic studies proved, many years ago, that most listeners, in a blind A/B test would say they preferred the louder version of otherwise identical recordings, even when only louder by tenths of a dB. So record producers wanted their music as loud as possible. In the days of discs, there were conflicting goals of how much would fit on the disc (and whether the playback needle would "skip" out of the groove) and how loud the disc was recorded at. Later, in the digital era the issue was digital clipping. On radio and television, it was controlling the peak modulation to avoid interference to other stations versus loudness. And advertisers wanted their spots to be heard above the already loud music. But the limits mentioned above were on instantaneous peak levels, whereas psycho-acoustic perceived loudness is much more related to average levels. So, people started compressing the dynamic range of material, music and advertisements, by using audio limiters and compressors. But on TV the commercials, where fidelity was the last thing on the advertisers' minds, ended up with much less dynamic range, and therefore louder, than the surrounding program. Audience members complained to the FCC with little action, and then to their congresspeople, which eventually resulted in the "Calm Act" forcing the FCC and television broadcasters (and cable companies) to act.

Now a little thought process on loudness. What makes something sound louder or softer? It all really comes down to the brains (and to some degree the ears) of the listener and what they consider loud or soft. If you hear someone screaming, even when the measured audio level (by any measurement method) is lower than the same voice not screaming, you may well say the screaming one is louder, because your mind is conditioned to think of screaming as loud. But beyond that there are many technical factors involved. How are you listening to it, what is the frequency response. "telephone" quality, or what level fidelity? How "flat" is the frequency response of the reproduction? Many people boost the low and high frequencies in listening well beyond the "flat" recording levels. If the loudest part of the material is in those lower or higher parts of the audio spectrum, those listeners will claim the material with those characteristics is louder than something with little bass or treble, despite the "loudness' or audio level of the source material measuring the same, And there are numerous other known reasons why material that measures equally will sound louder or softer in some cases than other material.

Now on to how we are measuring loudness. The first "loudness" meter was the VU meter, developed in the 1930s by ma Bell to measure levels on network program circuits. In that day and age it sort of worked, avoiding crosstalk between circuits and distortion in the transmitted audio. It was relatively easy to manufacture with accuracy at relatively low cost. It involves nothing but a meter with specified standards that was readily available, a built-in particular rectifier, and connection to the audio circuit in a particular manner. At the time, it did a fairly good job of matching audio levels (loudness) in most cases. Millions upon millions of VU meters have been made,

But, since it measures average levels, not peaks, it was no good for measuring modulation levels of radio or TV as the FCC required, the peak level that might cause the needle on a record player to skip, or in later years, cause clipping in a digital audio signal. Various peak or quasi-peak meters were developed, now often just software driving a bunch of LEDs or pixels on a screen. However, peak indicating meters do a lousy job of indicating loudness.

Back in the 1960s, as "fidelity" became more important to many, CBS Labs developed a "loudness meter" that was based on theoretical psycho-acoustic principles. It was generally considered to be quite accurate with most source material, probably considerably more accurate than the EBU or ATSC loudness meter standards. But it was expensive and never took off. Then CBS closed their labs, and the meter became unavailable.

For many years, TV audiences complained about loud commercials. Eventually the result was the "calm act". The FCC dumped the issue into the lap of the ATSC, a non-profit group of broadcasters and manufacturers which has developed the standards for over-the-air TV, looking for an answer. The ATSC ran studies looking for a solution that would "solve" the loud commercial problem (and incidentally solve the cable TV problem of different channels being at different levels). They wanted something that was relatively simple and not too expensive that would be simple for broadcasters to implement and would keep the FCC (and Congress) happy. They developed a standard numbered A/85 which was basically similar to the old VU meter with a bit of EQ in the meter circuit to somewhat better match the human ear's frequency response (which is actually quite complex and varies greatly between different people). Kind of a "quick and dirty" fix. The FCC adopted a requirement that broadcasters and cable companies follow it. At ABC we installed meters in control rooms, and software in editing computers to measure "loudness" to the standard, and protective limiters in our satellite and fiber outgoing circuits to control levels if the audio operators didn't. Other program originators and TV stations did likewise. That keeps the FCC happy.

Note, this was a US standard, not international, it did not apply to radio stations or recorded or streamed (on the Internet or otherwise) music or programs. The testing was done with typical TV programming, principally involving the dialog, not the music. (Read the standard, it talks about matching levels on the principle element of the mix, not the final mix. There was no intent by the ATSC or the FCC for it to be used to match levels of recorded or transmitted music.

After the Congress/FCC action, other national governments responded to similar complaints of loud commercials. But they developed their own standards, similar to A/85 but not identical. The most known version is EBU R128. As these were implemented, various shortcomings quickly became apparent. The most serious one was with programs with lomg periods of silence or very soft audio, common in dramas. These would often "fool" the algorithm, causing obviously incorrect measurements. The solution on later variants of the standards was to include "gating", where the measurement would be paused if the audio was below some level or silent. But the standards are only an approximation, still subject to error. And the FCC and Congress still receive complaints. But few people watch OTA TV, and less and less watch cable. Internet streaming is out of the purview of the FCC, so they pretty much have washed their hands of the issue.

Meanwhile, audiophiles became aware of the standards, and found applying them to music somewhat solved their issues with audio playback levels between media not matching (much of it being recorded with peak normalization, which is not at all a standard listening level). Since the standards were easy to adopt in software (like dBpa and some media players) they have become widely adopted (although with the recommended standard level adjusted to work with industry standards for the particular media, as it differs from a broadcast plant) , despite their imperfections. Also despite that the originators of the standards never intended them to be used for this purpose.. And they sort of work, to a degree. Certainly better than nothing. And the major Internet streaming companies have gotten on board, applying one or another of the standards to their feeds, not only long form material like drama and such, but also to podcasts and music. For better or worse.

Good luck, as I said, on developing something better. You might find it worthwhile to some of the many audiology studies on this subject (if you haven't already) and read the reports of the ATSC and EBU which resulted in the current standards.

Whatever "loudness controller" anyone develops, I'd be willing to bet I could find stuff to play that will "break" it. Your best bet is to choose something that is easy to apply and that is satisfactory most of the time. And if you find some media that breaks it badly enough, "fix" the media, either by adjusting the replay gain to play it at a loudness you like, or by adjusting the levels of the saved audio and/or using a compressor/limiter and saving the "fixed" result for playback.

John

[ForSerious]

Thanks for that John.
I think there are some pros and cons in what you detailed. One of my goals with my music collection is to normalize the volume as best I can. I use Stereo Tool to heavily repair all my music. Declipping, equalizing, pseudo-restoring high frequencies and a few more little things. So, for my collection, it might be easier to come up with a formula that can normalize the volumes, than in the wild-west of recordings that exist out there. Agreed, such a formula might be useless for the masses and easily broken—but I can hope that it turns out like that CBS Labs meter. I'm asking the people here at illustrate because they are more likely than I am to have the know-how to create such a thing.

[garym]

Good luck, but you should understand the history of the EBU (and equivalent ATSC) loudness measurements, why they were created and for what purpose, and the (serious) limitations in the measurement algorithms and in perceived loudness in general.
(...)

Thanks. We're lucky to have the expertise on this forum of schmidj