Tag
Investigates how self-supervised speech recognition models encode speaker group information (gender, age, dialect, ethnicity, native speaker status) across layers, and how finetuning for tasks like ASR or speaker identification affects this encoding.