We analyze a dataset providing the complete information on the effective plays of thousands of music listeners during several months. Our analysis confirms a number of properties previously highlighted by research based on interviews and questionnaires, but also uncover new statistical patterns, both at the individual and collective levels. In particular, we show that individuals follow common listening rhythms characterized by the same fluctuations, alternating heavy and light listening periods, and can be classified in four groups of similar sizes according to their temporal habits --- "early birds", "working hours listeners" , "evening listeners" and "night owls". We provide a detailed radioscopy of the listeners' interplay between repeated listening and discovery of new content. We show that different genres encourage different listening habits, from Classical or Jazz music with a more balanced listening among different songs, to Hip Hop and Dance with a more heterogeneous distribution of plays. Finally, we provide measures of how distant people are from each other in terms of common songs. In particular, we show that the number of songs S a DJ should play to a random audience of size N such that everyone hears at least one song he/she currently listens to, is of the form S ∼ N^α where the exponent depends on the music genre and is in the range [0.5, 0.8]. More generally, our results show that the recent access to virtually infinite catalogs of songs does not promote exploration for novelty, but that most users favor repetition of the same songs.