A sound description: Exploring prompt templates and class descriptions to enhance zero-shot audio classification

no code implementations • 19 Sep 2024 • Michel Olvera, Paraskevas Stamatiadis, Slim Essid

First, we find that the formatting of the prompts significantly affects performance so that simply prompting the models with properly formatted class labels performs competitively with optimized prompt templates and even prompt ensembling.

Audio Classification Contrastive Learning +2