How can I compare audio and display glasses fairly?
Give both formats the same task and score whether the wearer gets the information right. Include repeats, corrections and interruption, not only speed. Otherwise a fast but confusing answer can look better than a slower useful one.
Use information the participant does not already know
Use unfamiliar names or reference codes so memory cannot fill in a weak output. Keep the intended meaning equivalent between speech and text, and allow the controls each product normally provides. Record whether the participant asks for a repeat, misreads a character or abandons the interaction.
A sample task could be retrieving a made-up booking code and selecting the matching entry on a phone. Repeat it with different codes to avoid learning the answer. This is a suggested comparison method, not a published hardware benchmark; its value is making the errors visible enough to discuss.