Domain Adaptation Method and Modality Gap Impact in Audio-Text Models for Prototypical Sound Classification