Explainability is particularly necessary for speaker recognition because of its potential forensic applications. A recent approach, BA-LR, offers a new level of explainability. It represents a speech utterance by a binary vector where each coefficient indicates the presence or absence of a given speech attribute. However, the interpretation of these attributes remains unclear, as they are found automatically. The aim of this work is to develop a methodology for interpreting these attributes in terms of comprehensible explicit voice traits such as gender, age and perceived phonation type. Our methodology is based on the assumption that if an attribute is useful for a classification task of an explicit voice trait, this means that this attribute encodes all or part of this trait. Our results show that BA-LR attributes encode, at least partially, gender, age and perceived phonation type. These results pave the way for a comprehensive interpretation of BA-LR speech attributes.