Multi-lingual modelling of conversational speech

对话的语音识别仍然是一个非常具有挑战性的任务。然而，近年来，建模方法有了显着改进。语音技术面临的挑战之一是由于缺乏资源而覆盖世界上数千种语言。这个项目应该探索新的方法，可以利用在一种语言中获得的信息来识别另一种语言，特别关注符号方面的建模。如果这项工作取得成功，将产生巨大的社会效益。

Speech recognition of conversations remains a very challenging task. However in recent years modelling methods have improved dramatically. One of the challenges in speech technology is to provide coverage of the thousands of languages spoken in the world, due to lack of resources. This project should explore new methods that can make use of information obtained in one language to recognise another with particular focus on the modelling of the symbolic side. If successful this work will have significant societal benefit.