IASA 2017 Annual Conference

The session will present the use of computational tools to enhance discoverability of digital audiovisual collections. In the digital age, large A/V collections face the dilemma necessary cataloguing for search and discovery with search engines requiring text to retrieve relevant items. But as large numbers of analog collections are digitized, and with the influx of more digital materials, how can archives and libraries keep up? Computational tools are improving with the recognition of speech, images, and audio waveform patterns. Within the next five years, archives and libraries could begin to utilize these tools to automate much of the cataloguing of their digital files, or at the very least identify relevant people, topics, and locations associated with the content.

The American Archive of Public Broadcasting, with support from IMLS and in collaboration with Pop Up Archive, created more than 71,000 speech-to-text transcripts using a specially-trained version of the open source Kaldi software. We will show an online game called FixIt to crowdsource correction of speech-to-text transcripts.

The Netherlands Institute of Sound and Vision has been using automatic speaker labeling and thesaurus label extraction from subtitles to achieve fine-grained access. Currently we are at the stage of implementing 2.0 versions of these annotation techniques to improve quantity and quality. We intend to combine techniques and automate our workflows to enhance the results with minimal manual effort.

Brandeis University will demonstrate the suite of Natural Language processing tools available on the LAPPS grid. The Galaxy platform and workflow tools will also be demonstrated.