speech generation

Revisiting Automatic Voice Similarity Assessment: From Equal Error Rate to Embedding Geometry

Speaker verification models are often used as automatic proxies for human voice similarity judgments. This ana…