Loudspeakers radiate sound in multiple directions, not always equally. There are loudspeakers that radiate in very narrow beams, and there are those that radiate almost omnidirectionally. In-between these two extremes are designs such as dipoles that radiate in a roughly figure-eight pattern. All this depends on the frequency range in question, and loudspeakers have historically tended to get more directional as the frequency gets higher. Efforts in recent times have been towards engineering a more consistent directivity pattern over the audio frequency range, in order that the reflected sound in a room integrates well with the direct sound.
How, though, does loudspeaker directivity affect listeners, and is there a “right” answer to the question of the most appropriate directivity pattern for sound reproduction? Some of this seems to come down to what aspect of sound reproduction you want to optimise, whether timbre or spatial imaging. It also has to do with how much of a role the room takes in the spatial effect of reproduction. AES papers over the years have debated the issue, and have attempted to come up with evidence for one position or another. It partly seems to come down to the intended application and how many loudspeakers there can be. In sound reinforcement for example there may be good reasons for wanting to be able to control directivity to suit particular “aiming,” coverage or intelligibility requirements. In cinema sound there has been a tendency to want surround arrays to create a more diffuse or immersive effect, rather than localisable sounds, but that is changing with the advent of very large numbers of independent channels and object-based audio. There is then the question of what happens when you try to reproduce cinema-style effects in smaller rooms with only a few loudspeakers, and finally we come down to basic stereo reproduction of, say, music.