Music Should Be Played, Not Just Prompted
Type "an R&B song about grapes" into a generative AI tool and a finished song can appear in seconds. It is impressive. It also raises a fair question: if the AI does the composing, what does the musician do?
I found a thoughtful answer in a Georgia Tech research news story from this week.
An app that asks you to play
Gil Weinberg, a professor at the Georgia Tech School of Music, leads a lab that built HearMe, a web and mobile app for songwriting. His view is simple: "Music should be played, not prompted." So in HearMe, people play instruments, sing, and write their own lyrics. The app helps along the way.
For the grape song, it does not hand over finished lyrics. It offers line-by-line suggestions like synonyms, antonyms and rhymes, plus ideas for song structure. If you want to change a single word, it can change just that word and keep the vocal sound and accompaniment exactly the same.
Learning by making
The lab also designed the app to teach. Visualizations show pitch and song structure, and playful modes help non-musicians follow a beat. Weinberg, a lifelong musician who had never written lyrics, says the app encouraged him to try singing his own words, and he released an album made with its help.
Students in the lab are growing too. One undergraduate says working on it has deepened his wish to build music technology for a career.
Why I like this
As someone who builds machine learning models, I notice how often the goal is to remove the human from the loop. This project does the opposite. It treats AI as a patient coach and a good rhyming dictionary, while the person stays in charge of the ideas.
I think that is a hopeful direction. The best tools may be the ones that leave us more skilled, more curious and more proud of what we made.