This is an open-source music album based on ambient.garden, an explorable musical landscape. This album contains the same musical elements as the landscape, but instead of being interactive, it's written as a deterministic composition. Everything is generated from code without using any pre-existing audio samples.
Musically, I tried to escape sounds commonly associated with generative music. No AI models are used at any point, the composition is algorithmic but deliberate. The album is based on microtonal harmony and just-intonation, played by organic-sounding, physically modeled instruments which are heavily layered. The result is a deep style of ambient, with ethereal and classical influences.
You can hear the album on Spotify, Apple Music, and Bandcamp. Or you can build the album from code, the nerd way.
This repository has no audio files, but contains the full source code to re-create the album. The build script has 2 requirements:
- Deno, to run JS code. You can download it as a single binary or install it properly if you prefer.
- FFmpeg., for audio encoding.
To build the album, download or clone this repository, then run the following command in a terminal:
deno run -A render.js album
By default, this will create the album as mp3 files in the Generated/final/ directory. Note that this project does not really focus on efficiency (wouldn't be using Javascript otherwise!), the generation process needs about 1.5GB of RAM and takes 10+ minutes on a typical laptop. To customize album rendering, see the reference section below.
The music generation code was mostly written using the Teasynth web editor, which has features to play and debug music code, inline help for Faust library functions, etc. The MusicSource folder can be opened as a Teasynth project.
If you want to understand how any specific sound is made, start with the "macro" files in MusicSource/tracks/. These contain timestamped commands to start, stop and control "patches" within a track. The source for all these patches can be found in MusicSource/patches/. Javascript is used for high level composition code, while Faust is used for lower level audio-rate code. I recommend some familiarity with Faust before getting into .dsp files.
SUBCOMMAND: ALBUM
-----------------
Render the entire album, including cover art, to Generated/final/
Usage: deno run -A render.js album [--format=fmt] [--partial] [--ffargs ...]
Arguments:
--format=fmt final audio format (wav, mp3, flac...)
--partial keep existing rendered tracks
--ffargs ... all arguments after this will be passed to ffmpeg
Example 1: deno run -A render.js album
Uses default settings, rendering the album as 290k mp3.
Example 2: deno run -A render.js album --format=mp3 --ffargs -b:a 200k
Renders the album at a bitrate of 200k.
SUBCOMMAND: TRACK
-----------------
Render a single track to a wav file. Track names are in MusicSource/tracks/
Usage: deno run -A render.js track <trackname> <outfile.wav>
Example: deno run -A render.js track "06. occultation" test.wav
Renders the 6th track to test.wav
SUBCOMMAND: MASTER-TRACK
------------------------
Master a single audio file, applying dynamics compression.
Usage: deno run -A render.js master-track <infile.wav> <outfile.wav>
SUBCOMMAND: OVERLAP-ALBUM
-------------------------
Add overlap between separately rendered audio tracks.
Usage: deno run -A render.js overlap <indir> <stagedir> <outdir>
Example: deno run -A render.js overlap stage2/ stage3/ stage4/
SUBCOMMAND: ENCODE
------------------
Encode and add metadata to rendered tracks.
Usage: deno run -A render.js encode <indir> <outdir> [--format=fmt] [--ffargs ...]
Arguments:
--format=fmt final audio format (wav, mp3, flac...)
--ffargs ... all arguments after this will be passed to ffmpeg
Example: deno run -A render.js encode stage4/ final/ --ffargs -b:a 200k
Encodes the album at a bitrate of 200k.
SUBCOMMAND: COVER
-----------------
Generate the cover art to a png file.
Usage: deno run -A render.js cover <outfile> [--scale=N]
Example: deno run -A render.js cover test.png --scale=2
Generates the cover to test.png at 2x size.