Wildcat Voices: Remote recording using open web standards

Abstract

The University of Kentucky is conducting a speech production study with the goal of recording every member of our roughly 29,000 graduate, undergraduate, and professional students--the `Wildcat Voices' project: http://voices.uky.edu/. This talk describes the technology supporting this project, the strengths and shortcomings of this technology, and the challenges of crowd-sourcing a large database of recordings. The web site is written using open web standards including HTML5 Audio and Javascript. Participants can record using software already on their computer, tablet, or cell phone. This arrangement minimizes support costs and ensures interoperability with sites like Amazon's Mechanical Turk that do not allow software downloads. The site presents a prompt, records using the device's microphone, stores recordings in a database, and runs forced alignment to create a TextGrid object. Finally, this talk will present production data from the Wildcat Voices project and discuss the challenges of reaching all of the target participants, managing the server and backup infrastructure, and issues with recording quality that one does not face in a laboratory (computer fans, background noise, echo, etc.). Remote collection of speech production data is not without its problems, but the benefits of scalability offered by crowd-sourcing production studies are tremendous.