Artifact
An artifact is a sound a model introduced that was never in the original voice: a metallic ring, a warble, a click, a smear on a consonant. Artifacts are the giveaway that audio was generated or heavily processed rather than simply recorded.
Why it matters
Artifacts are the tell. A listener may not be able to name what is wrong, but a warble on a held vowel or a metallic edge on an s registers as not quite real, and it breaks the illusion faster than a flat delivery does. They are also cumulative: one is forgivable, a steady stream is fatiguing.
Often an artifact is not a synthesis problem at all but a cleanup problem, introduced by processing that was pushed too hard.
In practice
Audition on headphones, where artifacts are easiest to catch, and listen through held vowels and sibilants. If artifacts appear after a denoise or enhancement step, the fix is usually to back that step off, not to add another one.