Use the following HTML markup to reproduce the issue:

<div role="application" aria-label="Horizontal menubar">
    <div role="menubar">
        <a href="#" role="menuitem">Item 1</a>
        <a href="#" role="menuitem">Item 2</a>
    </div>
</div>

NVDA reads this markup in the following manner:

Item 1
Item 2

The expected pronounced state is the following:

Horizontal menubar
Item 1
Item 2

While the speech viewer tracks the aria-label="Horizontal menubar" word to its log, it does not pronounce it. Note that NVDA in Firefox does not have this issue.