Photo by cottonbro studio on Pexels

Can AI truly understand and enhance the nuances of the human voice? One experimenter put SUNO, an AI music generator, to the test by feeding it original acapella compositions. The challenge: transform raw vocals into fully realized songs. The subjects: a lighthearted tune about a self-driving car mishap, “I Got Hit By A Robo Taxi,” and a melancholic ballad, “Facing The End.” The user meticulously layered vocal tracks for the humorous piece and created a simple video for the ballad, setting the stage for the AI’s intervention.

Facing the inherent limitations of AI in orchestrating pure acapella, the user leveraged SUNO’s “extend” and “cover” functions, tools designed for manipulating existing audio. Segments of up to 120 seconds from each track were uploaded. For “Robo Taxi,” the entire song was used, while “Facing The End” used a segment consisting of a verse, chorus, bridge, and chorus. The process involved providing the AI with full lyrics and detailed style prompts, aiming to guide the creation of fitting musical arrangements.

The results were a fascinating mix of hits and misses. The AI sometimes misinterpreted the prompts, leading to music that strayed from the intended direction. In “Facing The End,” SUNO V4 occasionally mirrored the timing of the original vocals, even when it clashed with the AI-generated instrumentation. The AI also demonstrated a bias towards generating vocalists that sounded predominantly white, despite multiple requests for African American or R&B female lead singers. However, the user was surprised to see SUNO generate the entire song of “Facing The End”, which was a big improvement.

For those interested in diving deeper, the entire experiment, including the original acapella recordings and the AI-generated interpretations alongside their prompts, is available on YouTube: https://youtube.com/playlist?list=PL5OSzdNmlhc-PwVq6SKFCCP6o5DAUM2ZC&si=kHaK2K5BfchQynlK