The Vanderbilt University Libraries’ Vanderbilt Cloud Innovation Lab (VCIL) has released VAST, an open-source tool designed to help archives make large video collections more accessible to researchers and the public.
VAST (Video Autolabeling and Segmenting Tool) grew out of VCIL’s work with the Television News Archive (TVNA). What began as a solution to a specific archival workflow challenge has now been developed into a customizable open-source platform that other libraries, archives, and cultural heritage organizations can adopt for their own video collections.
The project represents an important next step for VCIL: moving from solving a local problem to creating reusable infrastructure for the broader archival community.
Making video archives searchable at scale
Video archives contain enormous amounts of information, but making that information discoverable is labor intensive.
Unlike a text document, a video does not naturally expose its internal structure to a search system. A single broadcast or recording may contain numerous stories, speakers, subjects, locations, and topics. For an archive to make those individual segments discoverable, someone must first identify where meaningful sections begin and end and then create useful descriptive information about them.
When collections contain thousands of hours of video and continue to grow, that work becomes a significant bottleneck. The result can be a backlog between acquiring video and making it meaningfully accessible for research. The original VCIL project was developed around the workflow and requirements of the Television News Archive to address this problem. The system uses automated processing to assist with identifying and labeling meaningful segments within television news broadcasts, reducing the amount of manual work required from archive staff.
The content-agnostic version of the project takes the underlying approach developed for the Television News Archive and turns it into a more general-purpose, customizable tool for video archivists. Instead of building the technology around the assumptions of a single collection, the open-source version is designed so that institutions can adapt it to their own archival material and workflows.
The result is a pathway for other organizations facing similar problems with large or growing video collections to experiment with automated segmentation and description without having to build an entire system from scratch.
For TVNA, the system is reducing staff workload and helping eliminate the video-processing backlog—the operational outcome the project was designed to achieve. Time previously consumed by repetitive processing can instead be directed toward the parts of archival work where professional expertise has the greatest value.
The Television News Archive provided a real-world environment in which VCIL could identify a specific problem, develop a computational solution, integrate it into an archival workflow, and evaluate whether the technology actually saved staff time.
VAST takes what was learned through that process and makes it available to others.
An archive adopting the open-source project does not need to have the same collection, metadata structure, or institutional requirements as TVNA. The goal of the generalized code base is customization: organizations can adapt the technology to the characteristics of their collections and the ways they want researchers and the public to discover their materials.
This is particularly significant for institutions with extensive audiovisual holdings. The challenge of providing access to video is not unique to television news. Oral histories, recorded lectures, performances, public meetings, educational recordings, historical broadcasts, and other audiovisual collections can all present similar problems of scale.
VAST provides a foundation that institutions can build on rather than requiring every archive to begin independently.
For an archive, the reason we keep what we decide to keep of the artifacts of the past is that they can tell us of our history, of our cultures, so access is pivotal. If tools allow an institution to process material faster, reduce backlogs, create more granular description, and expose portions of collections that otherwise would remain difficult to discover, then automation becomes part of the infrastructure connecting collections with the people who study them.
The technology developed for TVNA is already changing how its staff process incoming broadcasts. By releasing a customizable version as open-source software, VCIL is now making that work available to archivists, developers, digital humanities researchers, and cultural heritage organizations interested in applying the same approach to their own collections.
A project that began with a question about how to process television news more efficiently has therefore led to a broader solution.
How much more of our audiovisual cultural record could become accessible if archivists had tools capable of helping them describe it at scale?