[Updated 9/26/16 with a Reddit link on audio levels then, again, 9/30/16 with modified settings for web audio.]
For years, I’ve taught the following two audio concepts:
NOTE: When mixing audio for broadcast or theatrical release, your specs are much more stringent.
Both of those statements are still valid, but I’ve learned that they are no longer enough. We now need to also consider average audio levels.
BACKGROUND
Peak levels measure the instant by instant maximum level of a sound; peaks can apply to either an individual clip or a mix.
Average levels measure the level of a sound over time; which tends to reflect more accurately how the ear hears. While average levels can apply to a single clip or track, it is most appropriate for the entire mix.
Complicating the mixing issue is the relatively recent passage of regulations affecting sound mixing in both the EU and US; the Calm Act is the relevant legislation in the US. These regulations specify that, when mixing audio for broadcast or cable, average audio levels must not exceed -24 LKFS; where one LKFS is essentially the same as one dB.
While programs delivered to the web are not affected by these rules, unless they first appeared in a broadcast or cable program, paying attention to average levels can improve the quality of our mixes.
MY THINKING EVOLVES
I first started thinking about this three years ago as we were mixing the second season of the “2 Reel Guys.” We had the budget to work with a professional audio mixer for some of the shows and he started me thinking about mixing using average, rather than peak, levels. However, our studio sessions were short and I didn’t have much time to talk with him.
Moving into our video studio last year accelerated this, where our audio engineer – Ed Golya – and I would have long conversations about how to mix television programs. Ed is a multi-Emmy-Award-winning audio mixer and working with him every week was a treat.
Here’s a sample stereo waveform of an interview (left channel is on top). Notice how even the peaks are and how dense the waveform is? Peaks are around -6 and overall audio levels are consistently loud.
Here’s another example from a different interview. Here the peaks are also around -6, but there is a lot more variation within the clip. The loudness level will vary a lot as you listen to it. Peak levels alone do not adequately convey the difference in levels between these two clips.
We are all familiar with the audio meters, showing peaks on a moment-by-moment basis.
But, our mixes will improve if we also watch our average levels, as measured by loudness; such as this Loudness Radar in Adobe Audition.
NOTE: Here’s an article I wrote explaining how the Loudness Radar works.
UPDATE – 9/26
After I first published this article, Michael Kammes shared a Reddit link of professional audio engineers discussing where they set levels for broadcast and web. Read their views here. I found their comments on doing two mixes – one for broadcast and one for the web – so enlightening that it crystallized my thinking on settings for my own web work.
HERE’S WHAT I’M DOING NOW – 9/30
Here’s an example of an interview where things are out of whack. Notice the high peaks around 1:00, 2:30, 5:00 7:00 and 9:00, as indicated by the red arrows. These are sections where I am asking a question. (The softer passages are where the guest is speaking.) My mic is closer to me – and higher quality – than the guest’s voice over the phone. Thus, my levels are louder.
As you listen to the interview it is easy to hear these louder passages, even though the difference in peaks is not that great. (Yes, I’m using compression and limiting on the different tracks, which will account for some of this.)
NOTE: My past experience with poor quality computer speakers made me realize that louder is better – provided the audio does not distort. So, when I’m mixing for the web, I don’t bring my levels all the way down to -24. Instead, I use an average level hovering around -16 LKFS. However, broadcast/cable mixes must still follow their rules, which require -24 LKFS.
The Loudness Radar defaults to -24 LKFS; which assumes the audio output is for broadcast. For the web, we need to change this. Click the Settings tab and change the Target Loudness to -16 LKFS; this sets the second circle to match my web settings. This makes it much easier to see whether your audio is at the right level or not.
So now, after I’m done editing any project that has more than one audio source, I move that project – whether edited in Premiere or Final Cut – into Adobe Audition. There, I apply the Loudness Radar to the Master track so I can monitor my final mix. (The Loudness Radar is for monitoring only; it doesn’t change the mix itself.)
For example, this screen shot shows a much more even mix where no single section or speaker stands out from the rest. (The LKFS level is -17, all audio levels are relatively consistent and hovering around the second ring. This is a shade softer than -16, but close enough that I’d post it.)
Again, what I’m looking for is consistent audio levels which are as close to the second circle as possible, with an LKFS reading of about -16.
SUMMARY
As I said a the beginning, my opinions on audio levels are evolving. Peaks are important, but no longer enough. We need to watch our average levels. As I pay more attention to average levels, my mixes sound more consistent while the volume is still loud enough to be heard on a computer or mobile device speaker; i.e. devices with relatively poor quality and limited power.
Premiere and Audition include the Loudness Radar with their software. Final Cut Pro X has a variety of free and paid plug-ins that will monitor loudness. However, given the broadcast and cable requirements in both the US and Europe, loudness monitoring should be included with the software itself.
I’ll continue sharing my thoughts as I learn more and apply it to my work.
12 Responses to Rethinking Audio Levels [u]
Great article, Larry. You summed it perfectly with “peaks are important but no longer enough.” Logic’s latest update introduced a loudness meter. It’s not as colorful as the TC inside Audition, but it works in real time – which, at the moment, the Waves and IZotope plugins don’t.
Larry,
When using Loudness Radar or any other Loudness Meter that measures Integrated (aka program) Loudness, you want to define your Target in the meter’s preferences. In Loudness Radar, the “Radar” graphics represent the Short Term Loudness descriptor, using a 3 sec. time window. The second concentric circle in the Radar area represents the defined Loudness Target. Once you define your Integrated Loudness Target, Short Term Loudness may be accurately monitored during RT measurement.
The Integrated (Program) Loudness descriptor is the average Loudness over time. In order to establish an accurate reading, the audio segment must be measured in it’s entirety.
There is also a Momentary Loudness descriptor, indicated by the circular LED meter located above the Radar. The defined Integrated Loudness target is displayed at the 12 o’clock position. The time window for Momentary Loudness measurements is 400ms.
In terms of amplitude (peak ceiling) – it’s important to also understand the difference between Sample Peaks and Intersample Peaks. The latter are specified in all distribution specs.
For Internet/Mobile/Podcast/Screencasts. etc. -16.0 LUFS is a widely recommended Integrated Loudness standard, Of course I’m mainly referring to spoken word with or without video. Music production and elaborate soundtrack targeting have their own set of best practices.
In terms of True Peak compliance, it can be subjective, unless of course it is dictated in a given spec. For instance -1.0 dBTP for EBU R128, and -2.0 dBTP for ATSC/A 85. For Internet/Mobile I typically recommend no higher than -1.5 dBTP in the lossless file prior to encoding to lossy. This allows for any loss of headroom that may occur due to the lossy coding. In fact targeting -2.0 dBTP is even better. Again, somewhat subjective.
Bottom line – you are correct – Peak Amplitude has nothing to do with perception. It’s nothing more than signal level or a measurement of voltage. Use it to make sure the audio is not clipped. It’s the average Loudness over time *is* a direct indication of perception.
-paul.
@produceNewMedia
Paul:
This is a really excellent comment – thanks for the clarification.
Larry
Hi Larry,
Thanks for this article. However, when it comes to web it’s given me pause for something I have been doing.
A large ongoing project of mine is a series of educational video modules that are very simple by design: An talent speaking to camera–shot on green screen and keyed to white background with very clean, very simple motion graphics support in the white space. No other audio is used in these simple modules. Delivery is entirely via web portal and often viewed on institutional iPads and/or laptops.
The delivery format has been 1080p/30 MP4 files.
Owing to simple design of these projects, I focused on clean simple fundamentals. Recorded audio in studio with Schoeps mic to a target of -12dB, recording to 48khz as part of Prores 422HQ live capture. In post, knowing delivery was often to laptops and mobile devices, I felt we needed to maximize playback volume on these limited devices.
Therefore in post I gained up the clips, applied a soft amount of compression, and hard limited to -0.5dB on peaks (it only occasionally gets limited on transients). Overall the LKFS is about -12 with a loudness range of 3.8 according to Loudness Radar.
With over 500 finished and delivered modules out there, and our own internal testing indicating good results all seemed well…
But after reading this and the seeming consensus for -16 LKFS for web deliverables…now I’m wondering if I’ve pushed it too far. My gut says due to the simple nature of our content and our target deliverable we’ve done okay?
Carey:
This is a shade loud, but you should be OK. I try not to push that hard against 0 dB. On the other hand, music is frequently mixed that loud.
Here’s the basic rule: If it sounds OK, it is OK. For voice, I’d recommend the settings outlined in this article – music has different rules. But for your current projects, I wouldn’t worry.
Larry
Audio Levels Retunk !
Hi Larry ;
Would you consider a tut on using the tools currently in FCPX to achieve the best results = RADAR tool mentioned in the article ?
tomt
Tom:
At the moment, FCP X does not have any tools that equal the Loudness Radar. This is the main reason I haven’t written about it. I feel this is a significant limitation.
There are 3rd-party plug-ins, but I haven’t tried those, yet.
Larry
Hi larry-
Your article today stirred my interest to research this topic further:)
On YouTube I found this presentation followed by update by the same contributor .
Hi larry-
Your article today stirred my interest to research this topic further:)
On YouTube I found this presentation followed by update by the same contributor .
https://www.youtube.com/watch?v=ug5UQNg8F4k
&
https://www.youtube.com/watch?annotation_id=annotation_113612&feature=iv&src_vid=ug5UQNg8F4k&v=0EIRF7QW0eA
Two things are hghilighted here – the need to come to grips with “ Loudness” for broadcasting web material, and that there is one affordable ( $12 ) solution for the poor folks to 🙂
http://klanghelm.com/contents/products/VUMT/VUMT.php
Thanks for this Larry
Lastly this seems to be another useful tool to monitor the mean audio levels for $24 🙂
tom
Hi Larry – I thought I should share with you all .
I installed and tried this FCPX Plugin on one of my old finished projects and was so impressed that I wrote a blog about it 🙂
http://tomndi.blogspot.ca/2016/10/final-cut-pro-x-fcpx-lufs-level-metering.html
tomt
Just an FYI. Whenever I have a question I go to Larry! Your articles and videos are awesome and appreciated greatly by all of us unwashed masses in the editing realm. Thank you and thanks too to all who comment and clarify – a fantastic community.
Ashton:
Thanks for your note – it means a lot.
We are all happy to help.
Larry