A submission from ADAPT researchers took first place in the NTCIR-12 Spoken Query and Spoken Document Retrieval 2 (SpokenQuery&Doc-2) Slide-Group-Segment task.
SpokenQuery&Doc-2 forms part of the NTCIR-12 evaluation benchmark campaign, which focuses on search tasks in East Asian languages. The ADAPT Centre submission, which was developed by David Racca under the supervision of Professor Gareth Jones, will be presented at the NTCIR-12 conference in Tokyo and again in more detail at ACM SIGIR 2016 in July.
The SpokenQuery&Doc-2 task requires participants to use automated transcripts of spoken queries to retrieve relevant content from within automated transcripts of lectures. Both the queries and content are in Japanese. NTCIR was established in 1999 to advance the state-of-the-art in search technologies for East Asian language by offering challenging new tasks, rigorous comparative evaluation of participant submissions, and exchange of findings at the conference held at the end of each campaign.
The ADAPT Centre submission forms part of David Racca’s PhD research work which aims to develop novel methods to enhance the effectiveness of spoken content search. This work is exploring novel integration of audio features based on signal such as acoustic prosody, automated transcripts, semantic segmentation of transcripts and information retrieval models, to improve the accuracy and reliability of search over diverse types of spoken content.
Share this article: