Syllable-Based Sequence-to-Sequence Speech Recognition with the Transformer in Mandarin Chinese | Synapse