Mothers and fathers sang a song of their choice, once to their infant and once as if to their infant (simulated). The pitch of songs was higher and the tempo slower for infant-directed than for simulated versions. Listeners varying in child-care experience, musical background, and cultural origin reliably identified which of the two versions was infant-directed (Experiment 1). Identification accuracy was enhanced by musical training, knowledge of the singers' language and culture, and by greater pitch and tempo differences between versions. Other adult listeners who rated the singer's emotional engagement assigned significantly higher ratings to infant-directed than to simulated versions (Experiment 2). Differences in pitch and tempo between both versions predicted emotional engagement ratings. Finally, rating differences between infant-directed and simulated versions were highly correlated with identification accuracy.