A Few Observations about Grammatical Voice

I've noticed that some conlangers seem to think that grammatical voices basically only serve one purpose: omitting the subject. This somehow reduces the passive to serving as a convoluted way of saying 'someone did something to X'. 

Now, it would be lying to say that that is not part of what passives enable us to do, but a variety of languages use passives to a number of effects, and presenting a construction with an indefinite pronoun (i.e. 'someone') as the subject as a substitute seems to me to underestimate the amount of things that go on with the passive. Sure, it's a possible solution, and your language might well even parse the 'someone verbed X' construction as a legit passive with all the things that go along with a passive - the Germanic 'man' pronoun and the Finnish -taan passive are pretty close to that in some senses.

However, there are other uses for the passive beyond omitting the agent. In languages that are subject-prominent (as opposed to topic-prominent), the subject often has a very special pragmatic role. The subject, as it were, often correlates with some kind of 'topic of discourse'. So, a passive enables maintaining (and emphasizing) that particular relation: a patient that is felt to be important enough to be stated as the subject rather than object of its verb seems to convey the 'topic of discourse' thing pretty well - and study of how the passive is used would show that oftentimes, the subject of a passive verb indeed is topic not only of its clause, but of some part of the discourse.

Beyond this, we have languages with restrictions in their availability-hierarchies: comparing objects to objects might not be permitted (i.e. "I like sorbet more than (I like) parfait" would be forbidden, while "I like sorbet more than Sheryl (likes it)" would be possible) or, the distinction might be unclear due to fixed case marking on the standard of comparison, and passivization might help making it clear that the compared things are patients, rather than agents (in the case of more equal-status nouns, e.g. 'I like Edith Piaf more than Brigitte Bardot' - in this case, it might be unclear whether Brigitte Bardot is less liked or less liking!). More obviously, we have relativization-hierarchies: there are languages that permit relativization of subjects, but not of objects. Having a passive form helps with that, and other voices can help for things deeper in that hierarchy. 

A number of languages have verb forms that encode whether the subject is the same as that of the previous verb - so-called switch reference languages. Using the passive (or other voices) can permit for having same subjects for a longer stretch, since if some noun is the topic of discourse, it is likely to appear in most clauses - and probably in most of them as agent or other similar theta-role, but sometimes it might appear as a patient as well (and probably, in falling order of likelihood, various more oblique roles), and then voices can enable avoiding to have 'different subject' markers appear all that often, and the referent of the 'same subject' marker is kept constant, making it easier to parse an utterance.