Comparing supervised and self-supervised embedding for ExVo Multi-Task learning track