Narrating is a different job from performing. The voice has to keep out of the way for minutes at a stretch, carry the listener from one paragraph into the next and change gear when the story does. In AI narration, the failure I hear most often is a repeating melody. Each sentence starts high, dips through the middle and lands on the same falling note, until the pattern is louder than the words.
Good narrator voices break that pattern without being asked. Short sentences land a bit firmer. A question lifts at the end with no cue in the script.
You can hear this in a short sample if you listen to the joins. Ignore sentence one, which nearly every voice gets right, and pay attention to how sentence two begins. A fresh breath and a slightly different shape mean the voice can narrate. If it resets to exactly the same opening pitch, try it on a full paragraph before you commit, because that reset gets louder with every minute.