acoustic BPE

The article discusses the challenges in speech generation using discrete audio tokens derived from self-supervised learning models. It suggests that the current practice of directly utilizing audio tokens complicates sequence…