higher score for longer test utterance

i notice that when given short duration utterances, the PLDA score will have smaller aboslute value. Should there be something like score normalization?

The issue names no files, tests, or entry points. First clarify whether PLDA score normalization is expected and what behavior should be considered correct, then locate Kaldi's PLDA scoring path and identify tests covering utterance duration; done means an agreed normalization approach is implemented and its effect is verified.

Written by the indexing model from the issue text.