Programming Etude #1: "Poets of Sound and Time"

by Marise van Zyl

Description of Etude 1:

The first programming etude for Music and AI, which can be viewed here, focused on poetry. We were to use ChucK programming language and Word2Vec to help us create poetry that makes use of not only text, but sound and time as well. We had to create two very different poems. Here are mine:

Poem 1 - AiSMR

Created with AiSMR - a Poem creator and performer You provide it with a starting point and the number of lines that you want. For each line in the poem, it will then go on a random walk, choosing its favorite word out of the find_similar function and then use that as its new base word. I chose to feed the program very simple words as it provides something that is very very almost bad poetry. I also liked the idea of sonifying the thought process of the AI program as it processes these words and works through them. I recorded my voice whispering each letter of the alphabet. Using the words that the AI then chose, I went through each letter and ‘read them out loud’. The resulting product sounds like mutterings of the AI brain as it sorts through its words. The tool does not necessarily help with poetry, but it is wonderful to listen to - and isn’t that a form of poetry in itself?

no any not nothing there
no any not nothing there because without that if but
not be
be not because so
because even but not though could be still

yes maybe guess anyway imagine
imagine wonder maybe else happens anyway everyone surely
surely undoubtedly certainly

See the ChucK code for AiSMR here.
See AiSMR in action below.

Poem 2 - growing pains

In this poem, I sonify the results of a word similarity search and the results provides not only words to be used in each line of poetry, given a specific starting word, but it provides the FEEL of the poem. In a sense, this tool is more of an inspiring tool than one that gives you a direct poem. It makes the user think about the word or theme of their poem and ponder the accuracy of the sound that it gives. As I was playing around with different words and names, I put in ‘pain’. The resulting sound was very consonant and beautiful as I was using the 5 nearest similar words. However, I found that adding more and more words, the meaning of the words not only got more intense and sad, but the resulting sound became more dissonant. It made me think of how certain amounts of pain can be good, but that as soon as it becomes to much, the result can be unbearable.

In life, we have pain
Like sugar, small doses serve
Too much, we decay

See the ChucK code for growing pains here.
See growing pains in action below.


This etude was a tough one for me. It felt like I had to work REALLY hard to bring together the AI and the Poetry. The AI was a fun tool and although I still have no idea what word2vec is or does, it was fun to play with. It was also fun to play with Chuck (as per usual). It is in bringing the two together, that I felt was more effort than helpful. I usually love writing and reading poetry and I just couldn’t break the distance between me and word2vec. It seemed to get in the way of my creativity, instead of engaging it. The program that I created for growing pains, however, was really nice. I called it word2chord and played around with it a lot. It didn’t inspire poetry as much as it inspired conversation and exploration. I also showed it to some friends and they enjoyed playing with it. I wish the etude encouraged more interaction with the model itself or facilitated learning about AI, because, although I manipulated ChucK to produce ‘something’, I am still not sure exactly what I did.


Ge, YikAI, ChAI, Celeste, Bruno