Video Accessibility

We worked with Keith Gilbertson to come up with a way to make the videos on VTechWorks more accessible. Not accessible in the sense of where the videos are stored but accessible for searching for the videos and assistance while watching them. Initially, Keith wanted us to focus on both the ability to make the video full-text searchable and to add captions to assist with viewing. They wanted to make the videos easy to find and easy to watch if the video had tough accents or the viewer had a disability. He also suggested used CMU’s Sphinx software, a software that was developed to automatically transcribe videos to a text document. We also worked with Therese Walters of the VT Event Capture Group and Paul Mathers, a UNIX Systems Administrator with the library. Both of whom provided us valuable information and suggestions to move forward and complete this project.

After doing research on the currently available software for manual and automatic transcription and for captioning, we determined that captioning was far beyond the scope of this class project. Many of the automatic transcription tools we found were extremely poor at correctly identifying the words in a video and anything used for captioning was even worse. We even tested YouTube’s captioning and the accuracy was quite low. In the end we decided that working on a process for manual transcription and using Adobe Premiere for automatic transcription; captioning would have to wait for the technology to catch up.

For manual transcription we used a standard word document and wrote each word as we heard it. This process proved to take far longer than we anticipated for each video that was transcribed. The automatic transcription was run through Adobe Premiere and proved to be much more time consuming that we thought it would be as well. Typically, it seems to take just as much time to manually transcribe a video as it does to automatically transcribe a video. There appears to be a large separation between what a computer can process in the english language and what the human brain is able to interpret. There are also many other factors that go into transcribing a video, manually or automatically, which are discussed in further detail below.

In the end, there is no easy way to transcribe video, at least that is readily available. However, we believe the processes we researched can still be of great benefit to the library and the VTechWorks groups. We can further shift the focus of the project to focusing more on just the keywords than on transcribing the text word for word. This allows videos to be fully searchable and easily located on the VTechWorks site and can prevent matching the filler words; such as, the, and, it, and so on.

We want to thank Dr. Fox, Therese, Paul, and especially Keith for the guidance, support, and cookies given during the project. Below we explain in detail how to use and setup Adobe Premiere, the results of manual transcription research, and what we learned from the project.