Event Date
Wednesday, December 6th, 2023 – 12:00pm to 1:00pm
Speaker
Scientific Program Coordinator Charity Pitcher-Cooper
Abstract
YouDescribeX (the X is for eXtra eXperimental).
In this revised audio description tool for YouTube viewers can request AI described videos in addition to putting videos on the wishlist, the wishlist has been revitalized propting describers to make AD for those videos first, describers will have the option of using the freestyle version (watch the video, make a script, and record their own voice) or to have an AI supported version that auto captures all the text on screen, automatically chooses description track insertion sites by finding the gaps in the dialog, suggests possible audio description copy for correction and improvement by the describer, and then a synthetic voice reads the descriptions. Describers can choose to put their AD into a community editing process where their script and track placement is improved by other describes. I cannot wait to show you the good (wow is the text on screen capture incredibly good), the bad (the text on screen capture is so good it sometimes will pick up text on background things- like trashcans) and the ugly (descriptions for still images is still relatively poor, dynamic images are even more difficult) but our describers can correct the AI, and as the tool grows, their suggestions will lead to better audio description.
If you have an interest in the YouDescribe classic data (10 years of audio description data)! It can be found here: https://github.com/youdescribe-sfsu/You-Described-We-Archived
GitHub - youdescribe-sfsu/You-Described-We-Archived: This is the public repository to download from the You Described, We Archived dataset. This is the public repository to download from the You Described, We Archived dataset. - GitHub - youdescribe-sfsu/You-Described-We-Archived: This is the public repository to download from the You...github.com. In the future we will have three sets of data: auto described by AI, human edited from AI, and freestyle audio description.