Earl Nottingham got this idea started when he wrote:
I’m sure you’re familiar with the CALM act which sets new standards for how loud television broadcasts can be (especially commercials).
My question: What is the video creator’s responsibility (if any) to maintain specified volume levels in the final audio mix? The ATSC A/85 specifications of the CALM act mention a “new” amplitude reference level standard of –24 LKFS. I’m not sure how that relates to the VU scale.
This seems like something that will impact all video producers.
Larry replies: Actually, Eric, I have no clue; time to do some research. Here’s what I learned.
Loudness is more than a level on a meter, it is “primarily a psychological correlation of the physical intensity of the audio signal.” (Tektronix) Loudness is more than dynamic range. “Dynamic range” is the distance between the loudest and softest portion of an audio clip. “Loudness” is how loud a sound appears to the ear.
VU meters, which were used in analog audio, are notoriously inaccurate because of the ballistics in flipping that VU pointer across the dial. They tended to under-report peaks. Which, in an analog environment, is not a big deal – there was a lot of headroom. But, headroom doesn’t exist in digital. Digital meters display peak audio as dBFS (decibels full scale). We may call them “VU Meters,” but they aren’t.
Complicating this further is that peaks are not a good indication of loudness. As an example, think of listening to a couple whispering – while every so often we hear the loud tick of a clock. The clock is peaking the meter, but the couple is almost indistinguishable. This would not be a “loud” clip.
In a recent seminar, Tektronix reported that there is roughly a 15 dB range where audio levels are considered “acceptable.” Beyond that, listeners either want to turn levels up, or down.
Dolby is very active in trying to define, measure, and modify audio to create the optimum listening experience. This extends to broadcast where Dolby uses a variety of metadata in its AC3 audio streams in a broadcast network to help assure consistent audio levels. However, that metadata is only an option when listening to Dolby AC3 audio; which excludes any audio that is not encoded to AC3.
The problem is that as video editors, we need to start paying attention to these levels if we are doing the final mix for submission to broadcast. (If your audio will be mixed by a post-house, then this does not apply to you directly.)
Led by broadcast regulators in the EU, there is now increasing regulation on how loud is loud for things like commercials and programs in Europe and the US.
The ITU has created a spec – called ITU-R BS.1770 – that defines how to measure loudness on a consistent basis in keeping with these regulations. This standard creates a new loudness measurement: LKFS. (Soundtrack Pro does not measure to this value. Audition CS6 measures this value after the fact, but requires a plug-in to measure it in real-time during the mix. ProTools, too, requires a plug-in for real-time measurement.)
To see how this plays out in the real world, I contacted Woody Woodhall, CEO at Allied Post, an audio post facility in Los Angeles that does a lot of mixing for broadcast. I asked him how he measures loudness. Here’s Woody’s reply:
That measurement is made with a meter that is quite different than a simple VU meter. The technical [spec it needs to meet] is an LKFS measurement [using] an ITU-R BS.1770-1 algorithm; to read and mix correctly you must use a meter with those specs. Typically the Dolby meter, the LM100, gives their version of that reading. I myself use the Dolby Media Meter which can express this measurement. All TV that I mix conforms to this standard. I mix using three measuring devices to properly hit the network specs – a true peak meter, and RMS meter and a Dolby meter. QC [the quality control engineers at a broadcast outlet] can kick a program not within spec.
Now to further answer the question – if the program material being delivered must meet this spec for QC then the person in charge of creating and delivering must be sure to meet those specs. Be it a re-recording mixer like myself or a video editor who is creating the mix deliverable and splits.
There is a ton more geekery about it all but let’s just keep it simple…. The best software versions are the afore mentioned Dolby Media Meter 2 – MSRP $795 – great tool, provides a second by second log of the LKFS measure (great to send to QC when there is an issue) there is also a much more useful meter by TC Electronic – TR2 or TR6 – which can provide all the needed measures – true peak, RMS and ITU 1770. The Dolby meter can be used “faster than realtime” as an audio suite plug-in in Pro Tools or other DAWs. So you can create a log of the file after the fact. Create the mix and then run the file through the meter and a one-hour show can be measured in minutes. Of course that is just for verification… if you haven’t been measuring as you mix there is no way to tell if you are within spec.
One last gotcha… the spec is typically very tight, for my deliveries (this week!) for my series for Food Network, Fuse TV and truTV all the LKFS must be within 1 db, +/- of -24. Not much wiggle room anymore. -22 or -26 and you are kicked from QC. Gone are the days of “what sounds good is good enough.”
So, Eric, if you need to mix for something more critical than YouTube, now you’ve got your answer.
7 Responses to How Loud Is Loud?
Great article! It’s about time that we start getting serious about protecting peoples hearing and to reign in on wildly different levels from 1 source to another. It’s a good thing.
I would like to interject something else that I’ve never heard anyone speak of. There are millions of Americans (and I’m sure other countries too) that are hard of hearing for any number of reasons. Personally, I have a 90% hearing loss provided free of charge by Uncle Sam when I was in the service. The problem with these millions of people is that they have a “window” of hearing (range) that is much smaller than than “normal hearing” folks.
Specifically, too high a volume causes pain at a lower level than “normal hearing” folks can tolerate. This is the part of the new specs that will help protect ALL people when volume is loud.
The almost unthought of or known about part is when the volume is too low. We all know low volume does not “hurt” anyone. Hard of hearing people mostly miss any speech that is whispered.
I certainly realize the dramatic effect of having a very dynamic volume range. Producers and audio techs, however, should keep in mind that whispers (or any speech) that are close to the background level audio (within ~6db) are cutting out most if not all hard of hearing folks from being able to understand what is being said.
It’s called speech discrimination, meaning they can’t “understand” the words (discriminate them), because the complete “audio envelope” never reaches their brain where it gets decoded and “understood”. Bits get left out due to things like loss of high frequencies somewhere in the ear. Very much like trying to see fine visual detail in a film or program while wearing very dark sunglasses.
Whispers also reshape the audio envelope by changing: fricatives, sibilance, and plosives, which is an added burden for the hard of hearing. Not unlike wearing an eye patch on one eye underneath those sunglasses above.
The point is, those of us who have hearing problems also want to be able to hear & understand speech in your next academy award winner or nominee too. A good start would be to keep SPEECH within a 12 db range, and SPEECH at least 6 db over all background audio.
The intelligibility of speech in a mix is the result of the mixers taste and ability, the directors intentions, the playback situtation and sometimes the noisiness of the production sound.
Some mixers do prefer to have the whole dialogue mix be dynamic for theatrical mixes, but there is no general rule. I’m one of the mixers who wants speech to be understood when it’s necessary, which almost always.
What Woody stated above is one way to handle loudness normalization. I’ve never worked that way. When I mix to a spec that includes loudness normalization, I’m actually more concerned about the dynamic range they permit in terms of a maximum LRA(, not the final loudness, because that can be normalized across the entire mix or program segment.
The US btw is using the old spec. The ITU updated the spec to ITU-R BS.1770-2 over a year ago, though the ATSC will probably update their A/85 standard to match that at some point. The EBU R128 already did in March 2011.
The difference are two gates (-70 LUFS absolute and -10 LU relative). In reality those two specs, the A/85 and the ITU-R BS.1770-2/EBU R128 produce almost the same resulting loudness. The gates screen out silence and very quiet parts so the overall loudness value better represents the loudness the listener perceives across the entire program.
As for metering plugins, I’m surprised nobody mentioned NuGen Audio’s VisLM(live meter and audiosuite offline measurement plugin), LMB(batch processor) and LM-Correct(Audiosuite normalization plugin), as well as the Waves Loudness Meter plugin. They include full specs for not just the US but most other regions as well. The LM-Correct plugin in particular was made for Media Composer editors who can use Audiosuite plugins. This makes layback checks much easier for picture assistant.
The wiggle-room of +/-1dB is a bit weird I find. That this exists in the US must mean that some people cannot loudness-normalize after the mix. This provision does not exist in any other spec for delivered media, but only for live shows in the EBU R128 spec, where deviations are inevitable.
What a good, and I think underdone, topic Larry. The move to digital did not do audio metering any favours. At least in the early days. VU meters were literally there to show volume (volume units). That is they did not measure the level of the signal but the aparent loudness of it. You say they under-reported peaks Larry but I think that is a little unfair because they were specifically not supposed to report them. That was not their job. The BBC’s PPM was not dissimilar in its aim. It was designed to show volume too. However it also had a second function of displaying peaks. This gave it a fast rise time (for the peaks) and a slow fall time (for the volume). These did not have a response like a typical digital peak meter though the rise time was measured in milliseconds so that it compromised betwen peak and volume indications. The BBC PPM was there to try and stop over-modulation of am transmitters and then later analogue tape machines. In fact they were pretty well matched to analogue tape. To tell if you were over-modulating a transmitter, or tape machine, using a VU meter by contrast took lots of skill. The early digital meters really lost their way and tended to only be useful in reporting over-modulation of the digital medium. Typically they had instantaneous (one sample) peak response. This kind of indication has got nothing to do with audio volume.
You might be interested in this; many years ago I patented an audio meter designed for the digital age. This had a volume measuring indicator which would display a volume-type reading. A second indicator would then show peak levels on the same scale. There was a bar for volume and a line for peaks. These were driven separately though with the peak meter being a true peak measurement (not showing the peak of the other indicator as was somewhat common then). In this way the user could tell both how loud a peace was and whether it was likely to over-modulate. The two separate uses of meters combined in one display. The nice (IMHO) addition was that the peak indicator would not show at all unless you were getting near peak levels on the basis that if you are not close then it does not matter.
To measure the loudness the meter had a response curve modelled on typical human hearing so very high and low frequencies did not register at the same value as mid frequencies just as you would hope. One thing which became clear in my research was that an integration time of 250ms makes for a very good model of human hearing. Sounds that last less than a quarter of a second seem quieter the shorter they are but sounds longer than that appear to keep the same volume.
I wonder if better metering in the field, edit suite and elsewhere would help? Not the fancy analysis tools (which absolutely have their place) but properly designed everyday audio meters that show loudness like the VU tried to do. Certainly the BBC PPM, and I think to a lesser extent the VU meter, kept one in touch with how loud sounds were.
Hey great article! One question though. If I’m editing in Final Cut Pro or Avid, what’s the rough equivalent of -24 LKFS on the meters in those programs?
I can’t speak to Avid, but there is no equivalent in FCP X or FCP 7.
You’ve missed the point Chad, there is no equivalent scale as the metering method is a different concept, you’re not looking at peaks alone, you’re metering an average loudness calculated by the ITU-R BS.1770-2/EBU R128 spec.
There is no peak level equivalent of -24 LKFS — you never know where you’re at unless you actually measure LKFS. The LKFS measurement is frequency dependent and weighted so that certain frequencies tip it off more than others. A very sibilant mix will send you over quickly.
I recently had to re-mix a show that was “mixed” by a picture editor that didn’t pass QC. So, message to picture editors: Don’t do your own mixes and splits unless you really know what you’re doing. I’ve spent 15 years sound editing/mixing and still refining how I do things. For the average picture editor, it’s borderline impossible for you to create a mix that will pass network QC.