The research group of Professor Tatsuya Kawahara of the Graduate School of Informatics, Kyoto University, Professor Yoko Hirose of the University of Broadcasting, etc. is advancing research and development of automatic speech recognition for lectures and lectures, and by using the latest deep learning, the University of Broadcasting A recognition rate of approximately 90% was achieved for lectures as well.The OUJ uses it to create subtitles for online lessons because it can add subtitles more efficiently than transcribing them manually.This is the first time that this system has been used systematically.
The Disability Discrimination Elimination Law, which came into effect in 2016, requires that "necessary and rational consideration" be given to the removal of social barriers for persons with disabilities, and sign language and subtitles for the hearing impaired. It is necessary to guarantee information such as granting.In recent years, various lecture contents have been distributed on the Internet, but the current situation is that few have subtitles.Human and financial costs are also an issue.
Even at the largest media-based higher education institution in Japan, the University of Broadcasting, lectures on about 300 subjects are distributed on TV, radio, and the Internet, but currently, half of the TV lecture programs are subtitled. It stays at the degree.Therefore, this research group is advancing research and development on speech recognition and subtitles for lectures at broadcasting universities, introducing the latest deep learning using a large-scale lecture / lecture database, and further technical terms from textbook texts. By automatically registering expressions such as, we have achieved a recognition rate of approximately 90%.
In addition, as a result of comparing the case of editing the speech recognition result for about 30 lectures and the case of manually transcribing all the speech recognition results, an advantage is seen when the recognition rate of the system exceeds 87%, and when it reaches 93%. It was confirmed that the work time was shortened by more than 1/3.The developed system is being used to create subtitles for online lessons that started in 2016 at the Open University of Japan, and content with subtitles and still images added to radio lectures delivered via the Internet is also being experimentally distributed.In the future, it is expected that subtitles will be added to various lecture contents created by other educational institutions.
The research results were presented at the IPSJ Accessibility Study Group (SIG-AAC) on December 2016, 12.
