SpeechT5 is a unified-modal encoder-decoder pre-training framework that explores self-supervised...