Automatic Creation and Timing of Subtitles based on Sound Analysis
Date Issued
October 6, 2021
Type
Μεταπτυχιακή Διπλωματική Εργασία
Abstract
In this paper I explain the inner workings of IgniSub, a piece of software that attempts to generate subtitles given any audiovisual material. The software’s purpose is to create and sync subtitles based on various methods of sound analysis, namely voice activity detection, speaker diarization and parameter tweaking. Firstly, a machine learning algorithm is used to find regions of speech. Then, another algorithm further segments speech regions according to different speakers. Finally, subtitles are generated based on a scoring system which takes all factors into account.
Subjects