Humans use music for a wide variety of social functions: we sing to accompany dance, to soothe babies, to heal illness, to communicate love, and so on. Across animal taxa, vocalization forms are shaped by their functions, including in humans. In three experiments, we show that vocal music exhibits recurrent, distinct, and cross-culturally robust form-function relations detectable by listeners across the globe. In Experiment 1, internet users (N = 750) in 60 countries listened to brief excerpts of songs, rating each song's function on six dimensions (e.g., "used to soothe a baby"). Excerpts were drawn from a geographically-stratified pseudorandom sample of dance songs, lullabies, healing songs, and love songs recorded in 86 mostly small-scale societies, including hunter-gatherers, pastoralists, and subsistence farmers. Ratings demonstrated accurate and cross-culturally reliable inferences about song functions on the basis of song forms alone. In Experiment 2, internet users (N = 1000) in the United States and India rated contextual and musical features of each excerpt, which were predictive of Experiment 1 function ratings. In Experiment 3, we successfully replicated the findings of Experiment 1 in a small-scale population with limited exposure to Western music, finding high concordance between function ratings of internet users and Mentawai Islanders (N = 60). These findings are consistent with the existence of universal links between form and function in vocal music.