Tweet tweet

[jbum]

I've been wanting to do this using sonograms for Mockingbird songs. Guess I actually have to deliver now?

[jbum]

I noticed a few Moby Dick entries. Some years ago, I inserted the complete text of Moby Dick into an image, which you'll find here:

https://www.flickr.com/photos/krazydad/257804202/

[jbum]

So more on the mockingbird novel. The Northern Mockingbird is capable of singing continuously for hours on end. Individual birds have been estimated to have a vocabulary of hundreds of songs. The songs have an interesting hierarchical structure, organized into notes, songs and bouts, which are somewhat akin to syllables, words, and sentences.

The ideal goal would be to produce a book that contains the singing of a single bird, in sonogram form, but printed like fine ink & brush calligraphy -- the vanity memoir of a single bird. However, this will likely prove to be quite difficult in practice, because it's hard to get good clean sonograms from the recordings of a single bird, and it's hard to make the sonograms look good without a lot of labor, so I will probably have to find a more expedient solution.

[MichaelPaulukonis]

I get the feeling that, despite an entire month at our disposal, NaNoGenMo is all about the expedient solutions. We're creating shortcut generators. Valid choices all, just not... rigorous?

Again, not a bad thing -- there's a lot of shortcuts that show a lot of potential for adding to rigorous cases. And that's how the art develops, no?

You're a bit more on the side of I rigour, I note.

---

P.S. Any new names for "MOB brown" -- that strange orange-shift color that happens when photos (of people?) are composited together?

I keep seeing it happen, and I haven't found any literature on the subject other than yours.

[jbum]

Hi Michael,

I've taken to calling it "Emergent Orange" lately (or at least it's more saturated form). If you google that name you'll see some more recent activity -- I wrote up a paper summarizing my findings last year, and there was a piece about it in The Atlantic blog. I've gotten a few nibbles from researchers who are interested in figuring it out, but nothing more substantial yet.

[hugovk]

I've also had a play with similar averagings, plenty of emergent orange in The 7,063 Faces of 38,181 Magazines, but not so much in random Street View images.

I'll have to have a read through your paper and the Atlantic piece: http://krazydad.com/blog/2013/12/05/emergent-orange/ http://www.theatlantic.com/technology/archive/2014/08/the-color-of-every-photo-on-the-internet-blended-together-is-orange/378614/

[jbum]

Hi Hugo - nice pieces! You'll find the brown/orange crops up more when you use uncorrelated/randomly chosen photos.

[jbum]

So this is basically the effect I'm going for. I've made a bit of progress in that I have enough information to center the lines the way I want, by doing song/bout analysis based on temporal delays in the signal. I'm doing some post-processing of the sonograms to make them look a little more like ink.

One issue is word count. If we count a Mockingbird 'song' as a word, then I'm counting about 95 words per minute, and I would need about 10 hours of recordings to do a 50,000 word novel. If we treat a song as a sentence, then I only need an hour or two, however, I suspect I'm not going to be able to collect more than about 20 minutes of clean samples. So right now I'm thinking of treating this more like a book of poetry than a novel.

I think my page count will be closer to novel length (or novella length), but not my "word" count. I'm aware I could increase the word count by implementing a markhov chain, or making a synthetic song generator, but at this point, I'm inclined to keep it real.

[ikarth]

I think, given the process, keeping it real feels stronger than algorithmically extending it. I've been finding that a having a strong concept is one of the better ways to keep a work cohesive.

[MichaelPaulukonis]

(How) Did this turn out?