DPLAfest 2017

The American Archive of Public Broadcasting, with generous support from IMLS and in collaboration with Pop Up Archive, has created more than 71,000 transcripts of historic public broadcasting recordings using the open source Kaldi speech-to-text software. During this panel at DPLAfest, we will launch a game called FixIt to crowdsource correction of speech-to-text generated transcripts. Panelists will also discuss potential computational linguistic tools and methodologies to enhance discoverability of digital media collections. The session will demonstrate the results of 1) our work with HiPSTAS at University of Texas-Austin, who have conducted soundwave analysis and pattern recognition on a sampling of content in the archive, 2) the output and of the speech-to-text tools including name, topic and location recognition, as well as the implementation of this data to aid in search and discovery, and 3) methodologies and workflows around crowdsourcing the correction of transcripts.