Our paper on labeling audio using simple tasks and crowdsourcing has now been published in IEEE/ACM Transactions on Audio, Speech, and Language Processing and is openly available on IEEExplore. We divide the task of strong labeling into weak labeling of short audio segments that overlap, and then reconstruct the temporal information from the multiple annotator opinions and the overlap. In the process we also consider the competence of the annotators, estimated using MACE.
Journal publication on labeling audio
Strong Labeling of Sound Events Using Crowdsourced Weak Labels and Annotator Competence Estimation published in IEEE/ACM TASLP