Natural attentive man/machine conversation

Speech based dialogue with machines is now present in many areas of life, e.g. in call centres, information systems, voice dialling, to name a few. In all of these areas the conversation with the computer tries to satisfy an immediate need by the user for information that the machine can provide. That however is a simplistic view of the range of conversation options. In this project we will look at situations where the task is to bring new information into a conversation between humans, from events that are related to the topic of conversation, but that did occur somewhere else (e.g. in social media or news streams). Work in this area requires very high quality speech recognition for conversations as well as acceptable and socially adequate speech synthesis, but most importantly novel models and methods that represent conversational interaction and acceptable participation in conversations. Research into advanced highly adaptive learning algorithms, e.g. based on stochastic decision processes will be required.

The Speech and Hearing group at the Department of Computer Science at Sheffield University is devoted to the use of machine intelligence for natural interfaces and is known worldwide for its expertise in speech technology and bringing that to real world applications. Over many years the group has researched novel methods in speech technology and has provided software and services to a wider audience (e.g. juicer.amiproject.org, webasr.org). A wide range of resources, including extensive recording and usability testing facilities, high performance computing are available for research in this area.