9月24日，在2021中关村论坛之”人工智能与多学科协同创新”平行论坛上，北京智源人工智能研究院学术副院长、清华大学教授唐杰发布了”冬奥手语播报数字人系统”成果 图片来源/中国科学报 郑金武摄影
据国内统计，中国有2700多万听障人士，对教育、社交、娱乐、获取信息的需求巨大。如果算上需要与他们沟通的亲戚朋友和同事，涉及的人数将达数亿人。
[En]
According to domestic statistics, there are more than 27 million people with hearing impairment in China, and there is a huge demand for education, social interaction, entertainment and access to information. If we take into account the relatives, friends and colleagues who need to communicate with them, hundreds of millions of people will be involved.
传统的手语播报，主要靠手势来传递信息，一方面因不同的人有不同的手势习惯，容易造成信息传递偏差；另一方面，会造成手语播报主持人翻译工作量大、与节目主持人配合难度高，从一定程度上限制了听障人士对于信息的获取。手语播报AI数字人首先可以避免”千人千面”，始终保持统一；其次，可以根据收听到的语音，自动实时生成手语手势，提供精确的播报服务。
数字手语主播不仅需要一个真实的数字人体形象，还需要一个能够理解和翻译语音和手语的数字大脑。手语播音数字人是一种“智者”。它不仅表情生动、肢体语言丰富，还拥有一个能理解和翻译语音和手语的数字双轮驱动虚拟数字大脑。它配备了口型、表情、姿势、手部动作等素材，以进行“表情管理”。
[En]
Digital sign language anchors need not only a real digital human image, but also a digital brain that can understand and translate speech and sign language. Sign language broadcast digital man is an “intelligent man”. It not only has vivid expression and rich body language, but also has a “brain”-a digital two-wheel drive virtual digital brain that can understand and translate speech and sign language. It is equipped with mouth shape, expression, posture, hand movements and other materials for “expression management”.
2021年9月23日，北京，2021年中关村论坛上手语播报数字人 图片来源/视觉中国
为了提高冬奥播报的准确性，研发团队还对手语播报数字人进行了”培训”，构建了冬奥手语语料库，包括多模态肢体动作、表情、手指等语料，教授数字脑掌握手语表达的方法。同时，手语数字脑可以通过计算机模仿听障人士的大脑，将看到的中文文本信息转换成手语词汇序列，让听障人士实时了解赛事进程。通过AI技术将播报内容翻译合成为手语，并利用数字人虚拟主播来播报，成为解决这一难题的重要途径。
手语数字化播报之所以得以实现，是因为有了《武道2.0》超大规模人工智能模型的支撑。结合冬奥会相关信息，启迪大模型可以实现从文字到手势的自动转换。播音员播报时，数字人可以将文字转换成手语，方便听障人士观看赛事专题报道，这是《武道2.0》超大规模预训模式的首次实际应用。依托《数据》和《知识》，实现了冬奥会期间赛事新闻的手语实时翻译和播报，填补了该领域的空白。
[En]
Sign language digital broadcasting is realized because it is supported by the super-large-scale artificial intelligence model of “Wudao 2.0”. Combined with the relevant information of the Winter Olympic Games, the big model of enlightenment can realize the automatic conversion from text to gesture. When the announcer broadcasts, the digital man can convert the text into sign language, making it convenient for the hearing impaired to watch the special reports of the event, which is the first practical application of the super-large-scale pre-training model of “Wudao 2.0”. Relying on “data” and “knowledge”, real-time sign language translation and broadcasting of event news during the Winter Olympic Games has been realized, which fills the gap in this field.
2021年6月3日，观众在2021北京智源大会上参观智能模型”悟道2.0″的资料介绍。”悟道2.0″模型的参数规模达到1.75万亿，是目前中国首个、全球最大的万亿级模型 图片来源/视觉中国
据了解，这一”冬奥手语播报数字人系统”由北京市科委、北京市残联发起，科技冬奥专项支持，智谱AI、凌云光和北京广播电视台联合打造，项目还得到了市残联聋人协会的大力支持。
数字人的未来不仅仅是手语播报，随着人工智能技术的不断演进，数字人也进入了其他领域，比如向大家介绍空间站情况的数字记者小伟，游戏直播行业涌现的各种虚拟角色等等。
[En]
The future of digital man is not just sign language broadcasting, with the continuous evolution of artificial intelligence technology, digital man has also entered other fields, such as the digital journalist Xiaowei who introduced the situation of the space station to everyone, all kinds of virtual characters emerging in the game live broadcast industry, and so on.
如今，随着科技的蓬勃发展，数字人不仅达到了人像的现实主义水平，而且已经具备了一定的传播能力。虚拟交互技术的扩展让数字人可以进入医学、矿物学和航天等精细或危险的领域。未来，数字人将拥有与他人相似的视、听、说和知识逻辑能力，人工智能将得到进一步发展。让我们期待它的到来。
[En]
Nowadays, with the vigorous development of technology, digital people have not only reached the realistic level of portraits, but also have a certain ability of communication. The expansion of virtual interactive technology allows digital people to enter fine or dangerous fields such as medicine, mineralogy and astronautics. In the future, digital people will have the ability of seeing, listening, speaking and knowledge logic similar to others, and artificial intelligence will be further developed. Let’s look forward to it.
Original: https://blog.csdn.net/dmwh368/article/details/122471689
Author: 大漠无痕368
Title: “数字主播”上线：冬奥手语播报AI数字人
原创文章受到原创版权保护。转载请注明出处：https://www.johngo689.com/526898/
转载文章受原作者版权保护。转载请注明原作者出处！