We provide the outline of a (highly simplified) semantics for music. We take music cognition to be continuous with normal auditory cognition, and thus to deliver inferences about ‘virtual sources’ of the music (as in Bregman’s Auditory Scene Analysis). As a result, sound parameters that trigger inferences about sound sources in normal auditory cognition produce related ones in music – as is the case when decreasing loudness signals the end of a piece because the source is gradually losing energy, or moving away. But what is special about music is that it also triggers inferences on the basis of the movement of virtual sources in tonal pitch space, which has points of stability (e.g. a tonic chord), points of instability (e.g. dissonant chords), and relations of attractions among them (e.g. a dissonant chord tends to be resolved). In this way, gradual movement towards a point of tonal stability, as in a cadence, may also serve to signal the end of a piece, but on the basis of tonal information. The challenge is thus to develop a framework that aggregates inferences from normal auditory cognition and tonal inferences. We sketch a ‘bare bones’ version of such a framework, one that delivers a definition of ‘musical truth’.