An AI manual for Latin Music?
This week MudialMusica sent out a special edition of its publication SANDBOX, with a complete rundown of AI resources for musicians. MudialMusica is an organization dedicated to advancing the popularity and profitability of Latin music worldwide, and provides a wide range of resources to musicians throughout the Latin world. The edition is in PDF form, which is takes a few more steps to translate, but I skimmed it and saw that it is thorough – a kind of Music AI for Dummies.
I am not sure what to think of this development. I am not surprised – Mexico, and probably the rest of Latin America, is awash in computers, smartphones, and AI. So it is no surprise that one of the major Latin music organizations is on the bandwagon, especially given its partnership with Spotify.
Spotify’s growth in its user base in Latin America exploded in 2022 and 2023. This tracks with other streaming services, whose combined audiences made Latin America the fastest-growing music market in the world in 2021according to the International Federation of the Phonographic Industry.
So why does a Spotify partner put out a manual in Spanish for the use of AI in music, especially when AI may cause streaming services copyright headaches?
Well, lest we forget, Spotify was an initiator of AI in music. The first pop album created with the help of AI was Hello World, by Francois Pachet, a composer, and the director of the Spotify Creator Technology Research Lab. And things have taken off from there.
Among the many platforms available now to create AI-generated music are Authentic Artists, Bandlab , Infinite Album, Boomy, Harmonai and many more including ChatGPT ‘s Jukebox. There are also AI tools for creating music for films, TV, ads – etc.
So, tools for Latin music were inevitable but are they useful?
Latin music incorporates a high degree of human interaction. Yelps, cheers, claps, whistles- often thrown in at random. That can be programmed b y AI, but I suspect that the enthusiasm and excitement will not be there. It will sound like AI-generated bodice-ripper romance novels – they read like 50 every other romance novel but without the sex and excitement.
Latin pop and rock artists also incorporate different strands of traditional music into their songs – sometimes so subtly that gringo ears don’t catch them. And they often switch languages in mid-song, sometimes punning bilingually on the fly. I don’t see a generative language model pulling that off soon.
I think the heartbreak of a song like La Llorna is deeper than any LLM can produce. And I think the anguished singing in mariachi, boleros, and corridoes is beyond a model that simply scans what it has been fed and gives it back to in different forms.
I know what you are going to say – many of the corridoes and boleros cover the same ground that has been covered for 150 years or more. True, but the power of the song, not necessarily its specific lyrics, is what is important. And emotional power does not seem to be AI’s forte. Take a look at lyrics ChatGPT wrote for me at the prompt “please write lyrics for a heart-breaking bolero about a lost love”:
The guitar weeps with notes of desolation,
In this bitter rhythm of my heart’s creation.
(Chorus) Oh, eternal pain, finding no relief,
In the echo of sadness, I wander in grief.
My voice breaks as I sing your farewell,
The guitar weeps with notes of desolation,
In this bitter rhythm of my heart’s creation.
(Chorus) Oh, eternal pain, finding no relief,
In the echo of sadness, I wander in grief.
My voice breaks as I sing your farewell,
In this bitter rhythm of my heart’s creation.
Sort of the Taco Bell of songwriting and not a current threat to real boleros or corridoes. Maybe a different prompt could have improved it, but I am doubtful.
So, what is AI good for in Lain Music? Without a detailed translation (next week) it appears that one of the main AI uses featured in the SANDBOX document is in creating videos for songs. Given CGI and special effects, that I understand. More on this next week.
Patrick O’Heffernan
Illustration created by the AI program Dall-E2
Leave a ReplyCancel reply