Sometimes I have trouble sleeping, and this can have severe consequences the following day, especially if it is a workday. I figured that over the years, many techniques must have been devised for helping people fall asleep, especially in high-stakes settings such as professional sports, key value driving professions, the military, etc. So, I asked A.I., and it found me this technique, developed by famous U.S. track coach Lloyd Bud Winter, originally intended to help Olympic athletes relax, and later adopted by the U.S. military. It can, supposedly, make someone fall asleep within a couple of minutes.

The technique is quite simple, but it requires consistent daily practice to train your nervous system to flip the "sleep switch". Reports have allegedly shown that in the military, after 6 weeks of exercising this method daily, it has a success rate of around 96%. If that is true, it is phenomenal.

I created an adaptation of Bud Winter's technique, as a flat list of steps suitable for listening to as a narration when going to sleep. I call it an adaptation because I have removed some stuff that are unsuitable for do-as-you-hear narration, and I have added a few more steps at the end about additional sensory detail, which I saw proposed somewhere, but I do not know whether they exist in the original. I also moved the steps about relaxing the torso after the steps about relaxing the arms, because I found that in order to settle my torso I often have to use my arms, but then this may leave some of the muscles in my arms under tension, requiring the arms to be relaxed again.

Here it is:

• Close your eyes.
• Breathe slowly.
• Relax your forehead,
• your jaw,
• your tongue,
• and the muscles around your eyes.
• Let your shoulders drop as low as they can go,
• as if they are sinking into the bed.
• Relax your neck.
• Settle your torso.
• Take a deep breath,
• and exhale.
• Relax your stomach muscles
• and let go of any tension you are holding in your torso.
• Let your upper and lower arms go limp on one side,
• followed by the other.
• Relax your hands,
• and fingers.
• Relax your right thigh,
• then your right calf,
• right ankle,
• and right foot.
• Relax your left thigh,
• then your left calf,
• left ankle,
• and left foot.
• Your entire body should now feel heavy and limp.
• Now that your body is physically relaxed, stop thinking about any worries or tasks.
• Imagine yourself lying flat on your back in a canoe,
• on a completely calm, serene lake,
• with nothing but a clear blue sky above you.
• Embrace more sensory detail of your imagined scene,
• by bringing to your mind the smells of that setting:
• The cool humid air,
• the wet wood from the boat,
• soil, moss, and grass from the shoreline,
• green notes from water plants, and trees around the lake.

I have tried the technique a few times so far, and each time I am surprised to find that certain groups of muscles (for example, the muscles around my eyes) are in tension without me realizing it, and I only become aware of it when the voice brings my attention to those muscles, telling me to relax them. So, it seems like it does really help.

I created an mp3 file so that I can listen to it when going to sleep.

Here is version 1, there will be better versions:

Here is how I made it:

For converting the text to speech, I used the MAI (Microsoft AI) playground because it is free, and it is incredibly, astonishingly, mind-blowingly good. I experimented a bit and found that voice "Harper" with style "Softvoice" produced excellent results for my purposes. It does have a limit on the amount of text that it will process, so I had to do two prompts and receive two audio files.

For editing the audio, I used Audacity because it is a well-known, free, and decently capable audio editor. Here is what I did in terms of audio editing:

1. I very aggressively de-essed the speech.
   • After a lot of experimentation, I realized I had to use the Filter Curve EQ effect to cut out everything above about 5 kilohertz by configuring it as follows:
     1. From 20 Hz to 4 kHz a flat line at 0 dB, then
     2. from 4 kHz to 6.1 kHz a steep slope dropping from 0 dB down to -27 dB, and then
     3. from 6.1 kHz to 20 kHz a flat line at -27 dB (In retrospect, I may have de-essed it a bit too aggressively in version 1; I will try to fix this in version 2.)
2. I slowed down the speech.
   • Audio editing software have this awesome feature which allows us to alter the speed or the pitch of audio independently of one another; So, it is possible to slow down the audio without a corresponding drop in the pitch. I tried 75% speed and it was too slow, but 80% speed was just about right.
3. I inserted longer silences where I thought they were needed.
   • In order to give the listener plenty of time to follow each instruction before the next instruction starts.
4. I mixed it with the music. (See below.)

For the music, I used Google Gemini, because it is one of the few services that a) actually work, and b) do actually understand what I mean when I say "in the style of Brian Eno's ambient music". It only creates 30-second tracks, which is a joke, but for my purposes it was okay because ambient music can be repeated over and over again, especially when it is in the background with human voice in the foreground.

I decided that music is necessary as an introduction, in order to give me some time from the moment I press "play" on the bedside phone, to turn off the light, lie on the bed, and cover myself with the quilt before the instructions begin. In the beginning I was thinking of using several seconds of silence before the instructions begin, but I quickly realized that this would not work, because the problem with silence is that you never know whether you have actually started it, or whether you made a mistake and nothing is happening.

Finding an A.I. service that would generate the music according to my specifications, and for free, was the hardest part.

• Most generative A.I. services out there simply do not generate music.
• ChatGPT said that it could not do it, then it added that it was capable of doing it with some workarounds. However, when I asked it to try some of those workarounds I realized that it was only wasting my time. There is no way to get ChatGPT to generate music.
• Suno and MusicGPT (no affiliation with ChatGPT) did not understand the instructions and created audio that was not what I wanted.
• Udio demanded that I enable DRM on my browser, (I should have refused right then and there,) and once I did that, it went ahead and created some music which was decently close to what I had asked for, but it would, of course, not let me have it. How debauched is that?
• Soundraw did not give me any text box to describe what I want in text; instead, it follows a multiple choice quiz approach, which is, of course, so stupendously limited as to be completely useless.
• Only MAI (Microsoft AI) playground worked for me, although for most purposes, its 30-second limitation makes it as useless as the rest.

I will leave you with a quote from Lloyd Bud Winter:

Demand perfection of yourself and you will seldom attain it.
Fear of making a mistake is the biggest single cause of making one.
Relax: pursue excellence, not perfection.

— Lloyd (Bud) Winter