When using a screen reader and advancing to the next slide, the screen reader first speaks the contents of the previous slide before speaking the contents of that next slide. That means it takes time before you hear the new slide. And it doesn't seem useful to repeat the slide you just left.

Shower tells screen readers to speak the current slide by creating an element with aria-live=assertive and putting a copy of the current slide in it. However, it also sets the attribute aria-relevant=all on that element, which causes screen reader to speak not just the new content, but also the content that was just removed.

Shower should probably not set aria-relevant=all, but leave aria-relevant at its default value (aria-relevant="additions text", see https://www.w3.org/TR/wai-aria-1.2/#aria-relevant).

The code that sets this attribute appears to be on line 6 in core/lib/modules/a11y.js