CAU SUBMISSION TO DCASE 2021 TASK6: TRANSFORMER FOLLOWED BY TRANSFER LEARNING FOR AUDIO CAPTIONING
Abstract&Introduction&RelatedWork研究任务AAC(自动音频字幕)已有方法和相关工作面临挑战创新思路使用预训练模型,seq2seq模型使用CNN14和ResNet54作为encoder,transformer的decoder实验结论SPIDErscoreof0.246and0.285PROPOSEDMODELSystemOverviewPre-Processing输入