Developed and integrated a new audio identification scenario for the stanford-crfm/helm repository, expanding the HELM benchmark to support speaker recognition tasks using the VoxCeleb2 dataset. The work involved designing scenario logic for processing audio clips, comparing speakers, and implementing audio manipulation utilities to facilitate realistic evaluation of model performance. Leveraging Python and YAML, the developer focused on dataset integration and audio processing, enabling the benchmark to assess models on complex audio identification challenges. The contribution introduced reusable components for audio handling and provided a foundation for more comprehensive benchmarking coverage, with an emphasis on maintainable and extensible code structure.