端到端手语翻译研究
Research on End-to-End Sign Language Translation
杨淑莹 1田迪 2郭杨杨 2赵敏2
作者信息
- 1. 天津理工大学计算机科学与工程学院,天津 300384;学习型智能系统教育部工程研究中心,天津 300384
- 2. 天津理工大学计算机科学与工程学院,天津 300384
- 折叠
摘要
为进一步改善聋哑学生高等教育教学环境,提升聋哑学生人才培养质量,对语音到手语的专项翻译进行了研究.提出了一种新型语音到手语的翻译方法,直接建立语音到手语编码的端到端识别模式,再将手语编码序列匹配输出由基动作组成的仿真手语动画,省去了传统语音到手语翻译过程中的各个中间步骤,避免了其间的误差积累,提升了翻译效率及准确率.实验结果表明,该端到端语音手语翻译方法能够达到较好的翻译效果,在教学应用中达到了预期效果,具有一定的现实应用前景.
Abstract
In order to further improve the teaching environment of higher education and the quality of talent training for deaf-mute students,special translation of speech to sign language has been studied.A new speech-to-sign language translation method is proposed,which directly establishes the end-to-end recognition mode of speech-to-sign language coding,and then matches the sign language coding sequence to output simulated sign language animation composed of basic actions.This method eliminates all inter-mediate steps in the traditional speech-to-sign language translation process and avoids the accumulation of errors during the process,and improves translation efficiency and accuracy.Experiments show that this speech-to-sign language translation method can achieve better translation results.Also,it has achieved the expected effect in practical application of teaching,and has a certain prospect.
关键词
端到端/手语翻译/虚拟现实/深度学习Key words
end-to-end/sign language translation/virtual reality/deep learning引用本文复制引用
基金项目
天津市教育科学规划课题院教学成果奖重点培育项目(2019)(PYGJ-015)
天津市虚拟仿真实验教学项目(2019)(津教政办[2019]69号)
天津理工大学校级重点教学基金(2020)(ZD20-04)
出版年
2024
NETL
NSTL
万方数据