Your definition of soundness is incorrect. A theory is sound if it contains no sentences which are false with respect to a specific structure (or class of structures) of interest.
Now as an unfortunate matter of practice we generally only consider soundness when that specific context is understood in the background, so we omit it. For example, when we say
Peano arithmetic is sound
what we really mean is
Peano arithmetic is sound with respect to the structure $(\mathbb{N};+,\times)$; that is, for each axiom $\varphi$ of Peano arithmetic we have $(\mathbb{N};+,\times)\models\varphi$.
Meanwhile, Peano arithmetic is not sound with respect to the unique $1$-element $\{+,\times\}$-structure (basically, the trivial ring).
I think it's actually helpful to step a ways back and rephrase both soundness and consistency in a more general way. Specifically, given a class of structures $\mathbb{K}$ and a theory $T$, we say:
$T$ is sound with respect to $\mathbb{K}$ iff for every $\mathcal{A}\in\mathbb{K}$ we have $\mathcal{A}\models T$.
$T$ is semantically consistent with respect to $\mathbb{K}$ iff for some $\mathcal{A}\in\mathbb{K}$ we have $\mathcal{A}\models T$.
The fundamental difference then is the "every"/"some" distinction. There's a second difference too, however, which is down to practice: usually when we talk about soundness we're tacitly referring to a $\mathbb{K}$ consisting of a single structure, and almost always when we talk about semantic consistency we're tacitly referring to the class of all structures. So there are really two differences between soundness and semantic consistency at work here: soundness typically looks at a small context while semantic consistency typically looks at a large context, and with respect to that class soundness is a universal condition while consistency is an existential one.
That said, as I mentioned above we almost always work with reference to the class of all structures when thinking about semantic consistency. So in light of that your definition of semantic consistency is correct. Note that the completeness theorem shows that the "realist commitment" in the (with-respect-to-the-class-of-all-structures) notion of semantic consistency (namely, that mathematical structures are things which actually exist and about which sentences are true or false) can be removed; by contrast, there isn't really an analogous result for soundness(es).
