Construction and Summarization of Lecture Contents Using Both Slides and Lecture Speech

2014 Fiscal Year Final Research Report

• Project/Area Number: 23700115
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Media informatics/Database
• Research Institution: Toyohashi University of Technology
• Principal Investigator: TSUCHIYA Masatoshi 豊橋技術科学大学, 情報メディア基盤センター, 准教授 (70378256)
• Project Period (FY): 2011-04-28 – 2015-03-31
• Keywords: 自動要約

Outline of Final Research Achievements

Because lecture speech contains spoken phenomena such as filled pauses and silent pauses, a robust automatic speech recoginition method is necessary in order to realize automatic summarization of lecture speech. Our method consists of two steps: 1st step is to predict filler insertion locations and pause insertion locations against loosely transcribed corpora which has no pause information using filler insertion model and pause insertion model learned from precisely transcribed corpora including filler information and pause information, and 2nd step is to construct a language model based on both loosely transcribed corpora and predicted information.
And more, a method to detect lecture specific named entities was developed. The human annotation scheme to map lecture slides and lecture speech transcriptions was also established.

Free Research Field

自然言語処理