Embedded metadata is widely recognized for its ability to enhance the discovery, usability, and management of digital assets. This paper details the process of establishing and implementing core embedded metadata for a collection of over 23,000 digital audio master files at the United States' largest academic recorded voice repository. Obstacles and solutions discussed in this paper address both the common and exceptional issues in retrospective collection description, including idiosyncratic data sources, systematic metadata transformation, encoding inconsistencies, and file remediation.