How do you suppose an “8-hour track” gets made?
Perhaps rendered straight through for eight hours. It is not.
It is two and a half minutes, joined 258 times.
Why not render it straight through
Simply because it is impractical.
Synthesising eight hours in one pass balloons the computation and the file, and the smallest change means redoing all of it.
Every time I think “a little less reverb,” eight hours of processing would restart. You cannot iterate that way.
So instead: make a short seed, and build length by repetition.
- seed length: 150 seconds (2.5 minutes)
- repeated to: 8 hours
- which is roughly 258 times
Get the seed right and the eight hours are right. Fix the seed and everything is fixed.
The problem is the seam
This approach has one unavoidable weakness.
The joint between repeats.
Butt two pieces of audio together and the waveform jumps at that instant. The result is an audible click.
Over eight hours, that click happens 258 times.
A track that wakes a half-asleep listener 258 times. The worst possible outcome.
How the seam gets hidden
Three measures, layered.
1. Equal-power crossfade (8 seconds)
The end of one pass and the start of the next overlap and swap over eight seconds.
If you do this by simply adding the volumes, the sound dips exactly where they overlap.
The ear perceives loudness by energy, so adding two half-volume signals leaves a hollow in the middle.
So the curve used keeps energy constant through the overlap.
2. A natural sound over the joint
That still is not perfectly clean. So a quiet natural sound — waves, for instance — is laid across the seam.
The idea is not to erase the step, but to keep it from being noticed. Natural sounds already fluctuate, so they absorb small changes.
3. Verify by measurement
Finally, check the number.
How abruptly does the waveform jump across the joint?
Measured, that jump sits at about 0.012.
At that level, it does not register as a click.
The fade-in I forgot
The repeat routine included a fade-out at the end.
It did not include a fade-in at the start.
Which means the sound begins abruptly the instant you press play. For someone in a quiet room at night, that alone is a jolt.
A separate 10-second fade-in now sits at the head.
Obvious in hindsight — and exactly the kind of thing implementations drop.
Choosing the level
The other thing long-form demands care with is loudness.
Matching the usual streaming reference is a little too loud for sleep. These are meant to run quietly at night, so they are finished below that reference.
Playing a loud master quietly is not the same as playing a quiet master.
The former loses detail and turns muddy at low volume.
Why the seed is 2.5 minutes
Why 150 seconds?
Too short and the repetition becomes noticeable. A 30-second loop played 960 times will be recognised.
Too long and rebuilding gets heavy, so you iterate less.
Two and a half minutes sits below the threshold of noticing, while staying light enough to rebuild endlessly. It is a compromise.
Long-form tracks are not long because of a shortcut.
You build a short stretch carefully, then join it so no one notices.
More of my time goes into that joining than I expected.
Ideally the sleeping listener never notices once.
As long as it works, none of this is audible to anyone.
This article does not guarantee any medical or health benefit and is not a substitute for professional medical care. If you have any health concerns, please consult a physician or other qualified professional.
