Maria Eskevich - PhD Transfer Talk - 26th July 2012
Title: Towards Effective Retrieval of Spontaneous Conversational Spoken Content
Abstract: The volume of archived digital multimedia is growing rapidly with the expansion of technologies available for recording and storage. This material varies in topic, quality, source and form. Realizing the potential of this content requires that users are able to rapidly locate material relevant to their needs. While existing research has focused largely on search based on visual features for content where the relevant information is in the visual data stream, less attention has focused on data where the relevant information is predominantly in the spoken audio content. This presentation will introduce my research into the effectiveness of speech search and the development of technologies for search of challenging informally structured spoken content archives. The overview of spoken content retrieval begins with formally structured broadcast news and moves to more informal content in the form of meetings, interviews, and semi-professional recordings on the Internet.
My work concentrates on two types of data that consist of natural conversations and have real case search scenarios: meetings and Internet TV. My current experiments provide a deep understanding of the challenges and issues related to spoken content retrieval and indicate specific challenges which need to be addressed to improve retrieval eectiveness. The presentation will conclude with a plan for completion of my PhD.