This Week’s Program: Apr 4 - Apr 8

Weird week on the Internet. Productive week in the repository.

sonic-sketches is nearing a nice milestone. Let’s recap. I wanted to make a project where I could…

• Do something ambitious with the Clojure programming language. More ambitious then “Hello, World”.
• Learn to use the Clojure ecosystem at-large: Leiningen, CIDER, etc.
• Build something that wasn’t a web application.

sonic-sketches became that project where I could do the above and also work within a domain I hadn’t really ever explored before: music. As the project grew and expanded it went from an exercise in learning Overtone to the creation of an automated Bot that generated musical compositions.

Now, sonic-sketches produces a procedurally generated piece of music based on the day’s weather.

This week I put the “final” (nothing is final) touches on the song generation algorithm. Next week I’m going to focus on automating the bot: making it run at regular daily intervals without my involvement.

Here’s this week’s code.

96324d8387652758d1543fdf16314cdd8ce5ca58

From the Forecast API, I pull in three more data points: the probability of precipitation, the intensity of precipitation (in inches per hour), and the cloud coverage. I normalize these (for instance, making cloudy a boolean) and then map them to different intervals and music modes.

If the probability of precipitation is greater than 50%, heavy precipitation will make the song a minor scale. Medium precipitation will make the song minor pentatonic, and a light drizzle will choose the harmonic minor scale. I’m new to this music theory thing, but it is my feeling that those scales evoke the mood of a rainy day. If it’s not going to rain but it’s cloudy outside, the song will be in the Lydian scale which I think suggests something otherwordly or supernatural. If it’s a clear sunny day, we’ll hear the major scale.

Last week we used the day’s temperature to determine the pitch of the song and now we use precipitation to choose the scale of the song. With that, I feel like song generation is mostly complete for now.

28554e7b139693f26f9f3fe31b60249f23d91f40

One of the things I was doing in the REPL all the time was writing out this little piece of code:

(let [t (now)
      weather (:daily (:body (forecast/nyc-at t)))]
  (gen-song t (:data weather)))

I could change the value of t to replay a previously played seed. I decided to formalize this in the codebase and now I have a handly function to replay previously generated songs or generate a new one on demand. The difference between play-generated-song and gen-song is that the former will always make a request out to the Forecast API whereas the latter will just randomly generate values if the Forecast data is not provided.

efe9607fc73b4fdd54fe95cdf71f2f36bf104894

Here I wrangle out the day of the week from the RNG seed to use that as the first part of the song’s filename. I could have used clj-time here but I just preferred not to pull in the external dependency. This is the Java way of doing things. I don’t think I ever want to do Java outside of Clojure again. I much prefer working with Java in the context of Clojure. Later I move this into its own function.

fedac58e8cef07e7cc38aa91177cf16167e5070c

Back in song generation, I adjust the lead line (the 303) to play half notes and adjust the sequencing to account for that. I set the resonance control of the 303 to a random value. When generating the notes to play from the scale, I also randomly generate a value for the cutoff control of the 303’s lowpass filter for every note. This is like if you were to play a 303 line and just wiggle the knob as the sequence plays. It sounds more like a 303.

Here’s the song generated for today, Friday April 8.

The song plays at a moderate 96 bpm because the moon is slightly waxing crescent. The temperature today averages 45° and it’s clear outside so the song is in the key of E3 Major.

If you’re a musician or fan of music, I would love to know what feelings different keys and scales evoke for you.

🌚 Mark