A long while ago I posted a description of a voice, which I dubbed the 'dereflexive'. In retrospect, it is an unclear post. (Not an unusual problem on this blog, to be entirely frank.) So, let's try and rephrase the content.
Sometimes, in languages, you may have multiple possible third persons. Oftentimes, one is more prominent than the other, and will be the 'basic' third person you will assume a subject pronoun refers to. However, object pronouns in the presence of a third person subject often refers to a 'less' prominent third person.
I saw him
has him = prominent third person,but
he saw him
has he = prominent third person, him = less prominent third person.
Basically this sort of equates to something along the lines of proximal and distal. Now, it is not uncommon for languages to permit reflexivity by reflexive pronouns (or some other approach), and we thus get
he saw himselfvs.
he saw him
where 'himself' is the same person as 'he', and 'him' is a different person. Now, what if we can introduce a way of using the existence of this distinction in objects to distinguish the "semantic subject". Maybe by having a distal third person subject rendered as a vanilla third person direct object, but a proximal third person subject as a third person reflexive direct object.
So, some possible complications: maybe we want proper objects to still exist, and this we can permit by either demoting them to some kind of oblique position or maybe have double objects - if used strictly, this voice would only really be used with pronominal arguments anyway, so any regular noun will be "safe". Thus maybe we will have a two-pronged approach: regular nouns can be "regular objects" even in the presence of object-like pronouns with this voice, but pronominal objects have to be demoted to some type of obliques.