→ all versions of this pipeline
Five options for a quiet bed running under Shmili from beginning to end, each mixed under the same two minutes of episode 1 so the only thing that changes is the music. Every version here is matched to the same loudness, so nothing wins by being louder.
Read this first, because it might cancel the whole idea. Our own sound design notes already list “ambient beds under a story that is already intimate and quiet” as something that does not belong. That is this idea, named and turned down, in our own rules. Nothing measured here overturns it. What the measurements do add is narrower: if a bed goes in, the level is not the problem, the spectrum is, and one of these five is clearly the least bad.
A bed is not a fix for the handling noise. At a level that stays politely under speech, it covers three of the four ruffles in this excerpt and leaves the loudest one exactly as audible as it is today. It helps with the quiet tail, which is the part no detector ever found. It does not replace the work.
The one thing to listen for beyond taste: can you still hear the children clearly? The bed sits at the same level as their quietest moments. That is the real risk, not the music.
The same two minutes as it sounds today. Play it first, and again in the middle, or everything will start to sound like an improvement.
The only one where the 17-minute problem simply does not exist: it is one continuous hour of music, so we take a window and there is nothing to loop. Measured over the full render it shows no repetition at all. Steadiest of the five by a wide margin.
The catch. Just over a quarter of its energy sits in the same range as speech. And the genre it was filed under is, honestly, dark ambient.
Nobody owns it but us, so there is no permission to ask and nothing to renew across 81 episodes. It is also the gentlest of the five on quiet speech.
The catch. Two problems. Only a small part of it survives on a phone speaker, so at a whisper it may not read as music at all. And it is not as endless as I claimed: about every three minutes the texture measurably comes back around.
The kindest of all five to quiet speech, and it sounds like real instruments rather than a texture.
The catch. The terms covering what we may do with it changed once already between versions. That is a thin thing to build 81 episodes on.
The smoothest thing here, and the cleanest provenance of the borrowed ones: the musician uploaded it himself.
The catch. It is the one option that measurably rises ABOVE quiet speech in part of its range. That is the exact thing we said must not happen, so it is out.
Included so the boundary is audible rather than theoretical. This is what most podcast background music is like.
The catch. Most of it lives right on top of the voice, it has a pulse you will start counting, and it gives its loop away every five minutes. Play it once and the other four make sense.
Two minutes never reveals a loop. These are the two real contenders at full episode length, music only.