Reply to thread

Message: [QUOTE="Bombastus, post: 91985, member: 2178"] Several stems can remain below their individual peak ceilings and still create a higher peak when they are summed into the final mix. The reason is ordinary signal addition. Peaks from different files can arrive at the same moment, reinforce one another, and leave the mix bus with less headroom than any stem suggested on its own. A stem is not a miniature master. It represents part of the program, such as drums, vocals, music, dialogue, or effects, and its meter only describes that file in isolation. Once several stems meet at unity gain, their combined waveform is a new signal with its own sample peaks and true peaks. This is why a delivery rule such as “keep every stem below -1 dBTP” cannot guarantee a -1 dBTP sum. A ceiling applied independently to each file says nothing about how strongly those files will reinforce each other when they play together. [HEADING=2]Summing changes the peak structure immediately[/HEADING] Two identical signals aligned in time and polarity add directly. If each one peaks around -6 dBFS, their sum can reach roughly 0 dBFS before any extra bus processing occurs. Real stems are rarely identical, but the principle survives. Their instantaneous amplitudes combine sample by sample. Timing matters just as much as level. A kick transient in the music stem may land with an effects hit, a vocal consonant, or another percussion layer. None of those files needs to look dangerous alone. Their peaks only become a problem when the events coincide. Polarity and phase can move the result in either direction. Some frequencies reinforce while others partially cancel, so the summed peak is not something you can recover by adding the individual meter readings. Peak behavior can even change when relative polarity changes, an effect explored in [B][URL='https://aes2.org/publications/elibrary-page/?id=17662']cross-adaptive polarity switching for audio mixes[/URL][/B]. True peak adds another layer. The summed samples can reconstruct into inter-sample peaks that were not present at the same level in any single stem. Knowing [B][URL='https://goldmidi.com/community/resources/what-is-the-difference-between-dbfs-and-dbtp.70/']how dBFS and dBTP differ during reconstruction[/URL][/B] helps here, but the practical rule is simpler. Meter the combined output, not just the ingredients. [HEADING=2]Individual limiting can change the intended mix[/HEADING] Putting a true-peak limiter on every stem may look like an easy safeguard, yet it can create a different problem. Each limiter reacts to its own material without knowing what the other stems are doing. A drum stem may lose transient level while the vocal and music stems remain untouched, changing the balance that existed in the approved mix. The risk grows when stems are supposed to recombine into the printmaster. If the original stereo or surround mix passed through shared bus compression, saturation, clipping, or limiting, separately processed stems may not recreate the same result. Nonlinear processors respond to the combined signal, not to isolated pieces processed one at a time. A compressor is an obvious example. A loud snare in the music stem might pull the whole mix bus down for a moment, including vocals and effects. Process the stems independently and only the music stem receives that gain reduction. Add them afterward and the result no longer behaves like the approved full mix. This is also why universal stem ceilings are shaky advice. A stem intended for later creative mixing needs different headroom from a finished program master, while a broadcast or immersive delivery may come with explicit limits for beds, objects, rerenders, or recombined stems. The delivery specification wins over somebody’s favorite mastering number. [HEADING=2]The combined output is the file that must pass[/HEADING] A reliable stem check starts by importing every delivered stem into a clean session at the intended start time and gain. Disable accidental normalization, hidden pan-law changes, extra plugins, and automatic sample-rate conversion where possible. Then compare the recombined output with the approved reference mix and meter the sum. If the stems are meant to recreate the master exactly, a null test can expose differences that casual listening misses. Perfect cancellation is not always expected when the workflow includes lossy encoding, dithering, different render paths, or intentional alternate processing, but a large residual tells you the stem set is not simply rebuilding the approved mix. Peak compliance belongs at the point where the requirement applies. If a specification governs each stem separately, test each stem. If it governs the recombined program, test the recombined program. If both conditions appear in the delivery sheet, both need to pass. Leave room for whatever comes next. Summing, routing, downmixing, sample-rate conversion, encoding, and downstream limiting can all change peak behavior. A stem meter can tell you whether one file is safe on its own. Only the destination bus can tell you what happened when the whole package came back together. [/QUOTE]

Name