Issue link: https://digital.copcomm.com/i/1524736
18 FA L L 2 0 2 4 I C A S Q U A R T E R L Y sure that some very popular titles are in that 3%. I have delivered a mix to Netflix that exceeded that value that I thought played well, and there are nearfield mixes of major motion picture titles that play well in excess of this (maybe even up to 15 dB) that I personally enjoyed. So, again, all of this is highly subjective and contextual. Many other listeners almost certainly don't share my experience. So as far as advice goes, perhaps a value in excess of 8 dB can be considered a mistake… The other aspect of this issue is the importance of anchor- based loudness. A/85 (and therefore, the CALM Act in the US) clearly mandates loudness be anchored around the dialogue, despite widespread confusion and delivery specs that ask for full program measurements (which is in violation of the aforementioned CALM Act). The EU suggests full program loudness, but this is for a complex variety of reasons to do with the early listening of dialogue intelligence algorithms and the fact that a lot of European broadcasts are public, which means no commercials, and therefore, no obnoxiously loud interruptions to programming. They do, however, allow for anchor elements in a sort of "when appropriate for the content" way, which in my opinion should be "all the time." The reason I say this is that since people are primarily trying to understand the dialogue when they watch, they'll set the listening volume at the level that best achieves this. And if dialogue is not the anchor, then the listening level will change from program to program, and sometimes episode to episode of a given program. It just doesn't make sense to not use dialogue as the anchor element, so encouraging this is to our benefit. Our experiments with the Fraunhofer Listening Effort meter, along with the folks at Skywalker who have developed the CODA system (which can be used to create targeted mixes), have found that a lower delta value, while not directly tied to better dialogue intelligibility, can certainly be linked with increased subtitle use: The less the other elements compete with dialogue, the more you'll understand the first time it goes by, and the less you're going to "chase the volume" (a common colloquialism for needing to turn the volume up for dialogue and down for music and effects-driven passages). But this is of course, a trade-off with other creative considerations. MIXING CONSIDERATIONS So, how can we apply all this? We found targeting specific measurements can be very informative. As dry as that may sound, what it really comes down to is mixing with a tighter dynamic range than would be appropriate for a theatrical mix. Now I get it, we all love dynamics, and I'm not saying you can't have a dynamic nearfield mix, but in many cases, the full dynamic range of a theatrical mix is truly inappropriate for home listening. There is a mindset among some in the theatrical mixing community that the full dynamic range of the theatrical mix should be persevered for home listening if possible, and I feel comfortable stating that in many cases, this is a mistake. #SorryNotSorry A controlled PDLR of 8 LU is good for intelligibility, as is a target LRA for both the full mix and the dialogue. 20 dB is a decent outside guideline for full mix LRA, and a dialogue LRA of less than 12 or 14 is also reasonable. Taken into consideration together, these will all have a good effect on intelligibility, but I stress again that LRA should not be used as a requirement due to the way it is calculated, particularly with dialogue. And if you have genuine creative reasons to exceed these values, then you just have to use your judgment and blow through them. The recommendation is really just that you consider them carefully. There are also two new metrics that have been recently proposed by Fraunhofer IIS based on the BS.1770 framework. These are short-term (in the vein of momentary loudness) measurements designed to alert to potentially problematic passages in a given program. They work by considering the speech and other background signals separately, either from stems/ADM files or by using "AI" dialogue separation techniques on a composite mix. The proposed metrics are: SLD - Speech Loudness Deviation: The momentary speech loudness minus the integrated speech loudness. This shows areas where the dialogue deviates from the average by a given amount. SBLD - Speech-to-Background Loudness Difference: The momentary speech loudness minus the momentary background (M&E) loudness. This shows areas where the dialogue may be masked by background sounds. They aren't yet in practice, but may become available. Monitor level is another controllable factor that can help with real-world translation. We monitor nearfield at around 78 dBSPL (depending on room volume and target loudness spec) for a reason, but it can be helpful to lower the level and take note of what jumps out and what drops off. A swing in dialogue loudness will be a lot more perceptible at low level than at full reference volume (a level at which I feel I can confidently state almost nobody at home listens), and therefore, far better represents the real world. Mr. Mark Patterson puts it well: "Listen to the mix regularly in Dim (at least 10 dB lower) as it will reveal energy levels and what elements are falling below a threshold, causing the listener to raise the volume. If you force the listener to raise the volume in certain scenes, it will be too loud in others." Mix room size is another commonly ignored factor, which I