Methods for utilizing found speech data

My research is about investigating, producing and evaluating methods that utilize found speech data - speech that was not recorded with the purpose of being used in research (e.g. radio segments, public speeches, podcasts, archival recordings). Such data hold great worth in several regards but is typically challenging to work with. Contemporary machine learning approaches will only get you so far given the noisy and varied nature of these types of data collections, not to mention the scarcity of high-quality labels. Because of this, my research is mainly revolving around using a combination of methods such as unsupervised learning, visualization techniques, perceptual tricks with regard to human hearing and having an overall focus on including the human in the loop. Among other things, this has resulted in Edyson, a sound browsing tool that have shown promise not only in an exploratory fashion, but also as a viable scientific method used to utilize found speech. In my 80% seminar I will elaborate on these results and give a general report on my progress so far. Welcome!