People say that about the web, too. "If you write semantic html and use aria tags, it's accessible".
In theory, yes, but in practice it's probably a very bad experience unless actively worked on. For instance, I worked on a "seat chooser" once, that actually was very keyboard friendly and worked with screen readers. But trying to actually use it to select a specific seat for a blind was impossible. A seeing user can just press the seat directly, or tab quickly until the correct one is selected. But with voiceover it had to read out "seat 7B, $5 extra, not occupied, extra leg room, aisle, power outlet".
An actually useful way of selecting a seat would be to specify a seat directly, or some kind of filter for wanted features, a table or whatever. Not just take a visual way of presenting something and slap on some voice data.