More experiments with AI: choral music

Disclaimer for language teachers:  As you may have gathered from the title, this post isn't about languages, but it does include another language and the skills that I have picked up by using AI in my language work!

I have spent over thirty years being a language teacher, and over fifteen years writing about it on this blog.  During all of that time I have also been a choral singer and director.  I started choral singing in middle school (age about 9), continued at secondary school and the local youth choir, and then at university where I began conducting for the first time.  Since 1996 I have been a member of a local mixed voice choir (it was a condition of my first teaching job, but that's another story!)  I took over as conductor after our leader sadly passed away in 2020.  I'm always looking for new pieces to sing, and over the Christmas holidays undertook an experiment to create a new choral piece.

I decided (Why? asked one of my choir!) that I wanted the words to be in Latin.  Latin A level never leaves you!   I asked ChatGPT to create "a short 4 line poem in Latin about the joys of singing".  Its first attempt was a bit prolix so I asked it to simplify it.  Then it turned out that a four-line poem gets very repetitive in one piece of music, so I asked it to create a second verse.  Here is the whole thing:

Canto laetus sum
Vox mea clara est
Cantare amo
Cor gaudet semper

Canto cum amicis
Ridemus simul
Vox fortis, cor magnum
Gaudium manet

(I sing and I am happy
My voice is clear
I love to sing
My heart always rejoices

I sing with friends
We laugh together
Strong voice, big heart
Joy remains)

I changed the original parva (small) to fortis (strong).  Otherwise the words are exactly as ChatGPT created them.

Next I pasted these lyrics into Suno.com.  (Read more about my work using Suno here and here.)  My first prompt was "a capella, 4 part harmony".  This just gave me results that had clearly been scraped from the work of the vocal group Pentatonix.  I tried some other prompts, adding the adjectives happy and joyful, and then "in the style of a folk song", "in the style of a spiritual" and "in the style of Palestrina" among others.  Finally, after seventy-two iterations (yes - 72!) it finally gave me a song that I thought I could work with.

I downloaded the mp3 from Suno and transcribed the soprano (melody) and bass lines.  I then ran it through a free mp3->pdf transcriber, which didn't do a very good job but was enough to tell me the keys and possible chords quicker than listening and transcribing would.

The final step was to add in the alto and tenor lines.  I edited it a little to prevent it from being quite so repetitive, and added an ending phrase.

I introduced it to the choir last night, and they seem to like it!

So would I do it again?  Probably yes.  I have a new piece created especially for my choir, and it didn't actually take that long to get from the lyrics stage to the full score stage.  I used AI in the initial creation stage, but it still needed real person input to get to the finished piece.  It would doubtless have taken longer to start the piece completely from scratch.  However there is always the ethical and copyright concern of where the AIs have got the content from.