Dreamy Audio

2005-2006, 133 episodes

A podcast recorded after waking from dreams to recall what they were.

Episode #1 | Episode #76 | Episode list

Origins of the podcast

I was getting into the hype around podcasting, but I didn’t want to play nothing but music. I was at a loss for content. I’ve often been fascinated with dreams. My first website on Geocities was “Shoomi: Homepage of a Dreamer”. I had books on dreams that explored their symbolism. The idea of making a podcast about dreams was also a way to create unique content I didn’t have to plan in advance, and it gave me a great reason for an unstructured show. I have dreams every night. Sometimes many. They were weird and unique. Perhaps starting to pay more attention to them would help improve my memory or my dreams.

Recording

I would keep a digital voice recorder (DVR) by my bed. If I woke up at night, I would record what I just dreamed. I didn’t have to turn on a light and write in a dream journal, so you get the raw state that I was in before I went back to sleep. In some instances, I’m commuting between work and home, talking about what I had dreamed.

Olympus W-10 DVR

I started out with an Olympus W-10, purchased on April 12, 2005, on Amazon, for $39.87, with a built-in microphone, 3 hours of voice recording, and 16 MB of memory that could store up to 100 WAV files. In high-quality mode, I could only get 45 minutes.

Olympus WS-200S DVR

A year later, around May 3, 2006, I purchased an Olympus WS-200S on eBay for $100 ($30 less than retail). It had 128 MB of storage for up to 55 hours of speech (about 9 hours at 44 kHz mono), had 5 folders to store 199 files each, was powered by a AAA battery (~13 hours), connected directly to a USB port without a proprietary cable, and recorded in stereo. It connected to the computer as a USB drive. It was also easier to use in the dark without turning on a light: press the top record button twice to turn on and record/pause, and press the button below to stop. As of January 2026, the WS-200S still works. [instructions as PDF]

Olympus ME-12 Microphone

I was still having trouble with the audio quality. On July 17, 2006, I purchased a Noise-Cancellation microphone (ME-12) from Olympus for $16.95.

Recovered Assets

Image Artifacts

The dominating resolution at the time was 640×480, so the banner displayed at the top was designed to fit within that range.

The 80×15 site badge was common at the time and appeared on most websites to brand the site or the services it offered or made available. An icon file (Favicon.ico) was used to display the favorite icon in some browsers when the page was bookmarked. The icon for my site was a grey speaker icon with two yellow Zs stacked on top of each other, pointing to the right. You can store both a 16×16 and a 32×32-pixel icon in the same ICO file. Here, they have been exported as two separate GIF files. The larger one is pixelated because it was quadrupled in size without interpolation. Some browsers may support animated GIFs as icons, so I created one that displays the standard 16×16 icon and then shows a marquee reading “Dreamy Audio” – a serious flex in those days. The 64×64 icon was used in the RSS feed to identify the podcast, which was the standard at the time. It was a photo of a small green dreamcatcher that I had over my bed, with light green beads and green fluffy feathers, hung against a spring green-painted wall.

Metadata from MP3

Rather than using a database or a web page to store metadata for each episode’s name, number, date, image, and comments, PHP files were created to read the ID3v1 and ID3v2 tags embedded in the files. By dropping an MP3 file into the directory, the website immediately provided visitors with information, including Atom and RSS feeds, extracted from the MP3 file’s metadata. The file itself is a single source of truth and would carry all information when downloaded, so if people archived it, they would have all the original information I had.

Lost Dates

Not everything was preserved. As time progressed, not everything was stored in the MP3 files – or at least, not in the ones I have on hand. Date and times for early episodes were found on the old blog at http://dreamyaudio.blogspot.com, which is still semi-live today. MP3 Files are all broken links to AudioBlog.com. Episode 21 was also hosted on Odeo – the inventors of Twitter (now X) as an internal communication tool.

However, the last episode listed is #30 on July 18, 2005. The MP3 file for episode #31 contains the date October 7, 2005, at 5:59 AM in its ID3v1 tags. After that, I have no dates within the MP3 files. It appears I relied on the file system’s creation dates, which were not preserved when I checked my USB drive.

The Internet Archive’s Wayback Machine has a few snapshots of the front page that helped me anchor the latest possible dates for episodes 93 and 131.

• May 11, 2006 snapshot: shows episode #93.
• February 8, 2007 snapshot: shows episode #131

Some of the RSS, Atom, and XSPF feeds were also archived. Unicode in #126 broke part of the RSS feed, but I was able to capture the missing information in the Atom feed from the same day. Supporting multiple feed formats proved unexpectedly valuable. I just wish I listed more than 10 episodes – but that was the standard back then.

• #84 Fri, 5 May 2006 23:45:26 -0400
• #85 Sat, 6 May 2006 00:00:20 -0400
• #86 Sat, 6 May 2006 10:58:30 -0400
• #87 Sat, 6 May 2006 20:50:09 -0400
• #88 Sun, 7 May 2006 21:30:07 -0400
• #89 Sun, 7 May 2006 21:30:50 -0400
• #90 Mon, 8 May 2006 21:49:50 -0400
• #91 Mon, 8 May 2006 22:08:55 -0400
• #92 Thu, 11 May 2006 01:50:45 -0400
• #93 Thu, 11 May 2006 02:17:51 -0400
• #122 2006-08-01T00:09:09-04:00
• #123 2006-08-02T00:41:48-04:00
• #124 2006-08-03T03:31:59-04:00
• #125 2006-08-04T00:43:21-04:00
• #126 2006-08-06T01:15:31-04:00
• #127 Mon, 7 Aug 2006 02:05:49 -0400
• #128 Tue, 8 Aug 2006 23:22:06 -0400
• #129 Fri, 11 Aug 2006 01:25:45 -0400
• #130 Mon, 14 Aug 2006 01:48:24 -0400
• #131 Mon, 14 Aug 2006 02:18:38 -0400

I tried looking up my old podcast ID on iTunes, but it just threw an error after a minute, and the Wayback Machine doesn’t have a cached version of the page.

For the remaining episodes, I’m using documented circa date ranges based on the closest available archival evidence.

XSPF

Wait… did you say XSPF? I sure did. While RSS was competing with Atom, I was also using another format with a Shockwave Flash player on the website. It was compatible with a simpler format called XML Shareable Playlist Format (XSPF), pronounced “spiff.”

It doesn’t have anything like the date for each episode, as its focus is primarily for playing one song after another in a music player. Here is an example of just one track:

<?xml version="1.0" encoding="UTF-8"?>
<playlist version="0" xmlns="http://xspf.org/ns/0/">
  <title>Dreamy Audio</title>
  <creator>Lewis Moten</creator>
  <info>https://lewismoten.com/</info>
  <location>https://lewismoten.com/xspf</location>
  <image>https://lewismoten.com/logo.png</image>
  <link>https://lewismoten.com/podcasts/</link>
  <date>2007-08-19T00:41:43-04:00</date>
  <trackList>
    <track>
      <location>https://lewismoten.com/131.mp3</location>
      <image>https://lewismoten.com/131.png</image>
      <annotation>Dreamy Audio is a collection of my dreams.  I usually record them on a small audio device when I'm half-awake.  I may analize my dreams, or just explain what I have seen.  The real world may influence my dreams from time to time. Please keep in mind that the content of my dreams does not represent my own opinions, thoughts, or what I have done in the past.</annotation>
      <title>131 - Covered Mouth</title
      <info>https://lewismoten.com/131.htm</info>
      <album>Dreamy Audio</album>
      <duration>739161</duration>
      <trackNum>0</trackNum>
      <creator>Lewis Moten</creator>
    </track>
  </trackList>
</playlist>

Tools from the future

Speech to Text

Eventually, I started transcribing the episodes myself back in 2005 and embedding the lyrics in the MP3 files. It became a long, drawn-out process. I tried using some free speech-to-text services, but they were very poor, and my audio quality was poor as well.

In January 2026, with advances in artificial intelligence and widespread adoption, services were available to address this. I used Revoldiv to transcribe the audio into a text file. I would then include that text in a hidden section of the page that would reveal itself if the user clicked the “transcript (auto-generated)” link. This, in turn, made the MP3 files’ content searchable.

Summarizing with LLM

Afterwards, I would put the text into a large language model, with a prompt similar to the following, changing the episode number and pasting in the transcript at the end:

What I got was quite a lot of content and metadata to assign to each episode, which could explain what the dream was about, display a short summary when listing the episode on other pages and in search results on the blog, and link to other dreams using tags. I gave each episode a new title that better reflected its content, whereas my original episodes were 1-3 words and very general. I still preserved details in a heading block on each episode’s page, such as the original date published, and date restored.

Dream Analysis and Related Dreams

As I continued processing each episode, I began modifying the prompt. First, I added a dream analysis. It’s quite interesting to see the interpretation of dreams from the perspective of a cognitive system with substantial processing resources. I’d never have the time to read through that much material in my lifetime. Since I was telling the specific episode number I was working on in my prompt each time, this gave me an opportunity I hadn’t planned on. I started asking it to tell me how it was related to past dreams. Not only did it specify the episode numbers, but it also gave a brief summary of the elements in those dreams that matched the current dream. It didn’t specify the titles, as I had previously asked for a suggestion rather than a single title. It also started identifying arcs and patterns.

One thing I’ve noticed, though, is that as I go through each episode, the reference to other dreams changes. It stopped mentioning themes in parentheses and stated that the specific dream number shares specific symbols and such.

I don’t have exact dates, as I’m around episode 50, but it seems college must have started around this time. I started having dreams of both high school and college, test anxiety, and forming opinions about seeking credentials when you are already very good in your profession.

Image Generation

Next, I asked it to describe images to display next to each paragraph in the summary, as well as the post’s feature image. For an added bonus, I asked for the file names as well, in the format I often save them in. I would hop over to DeepAI and paste in the text and use the Dreamcape Generator as the style, and just use the lowest model for what used to be high definition before Genius and Super Genius came along. The lowest model wouldn’t burn through my credits as fast, and I would actually get to use up my credits this month rather than letting them go to waste.

Once I added the images to the page, I pasted in the image descriptions into the image alt tags.

Similar Dreams in History

As I continued asking the LLM to analyze and link to related dreams with image descriptions, curiosity kept getting the better of me. Had anyone else had similar dreams? Part of it was a reach for feeling “normal,” and the other was trying to gauge how common the dream was.

I expected maybe an occasional celebrity who mentioned similar topics, but instead I was often given information about the psychologist Carl Jung and the author Franz Kafka, their patients, and their works. I tried to encourage more variety by avoiding the two.

Restoration Workflow Video

The following is a snapshot of my workflow during episode 67 – the official halfway point of 133 episodes. Speeding along, I got everything set up in about 6 minutes, with image generation and uploading taking the longest. With 66 episodes remaining, I had about 6.5 hours of work left if I kept focus, avoided distractions, and nothing changed.

Plugin Development

Speaking of distractions, I couldn’t find a plugin that would generate a tag cloud filtered by category. The cognitive system kept insisting the page needed one, so I decided if it wanted it badly enough, it could help build it. I started vive coding—AI-assisted development in real time—and let it do the grunt work. Eight hours later, after many corrections, I had written my first WordPress plugin with a few extra features I wanted.

Series Taxonomy

Throughout each dream, various tags have been assigned to represent its themes. The following is a tag cloud, where larger tags indicate more frequent themes. As more episodes are added, you may see eras, arcs, and motifs appear.

Episodes

This project is not a remaster or reinterpretation. Original audio remains unchanged. New titles, summaries, and images are additive, created to improve discovery, accessibility, and context while preserving the original recordings intact.