This paper is intended for an audience of speech technology specialists who believe that "automatic processing of under-resourced languages is a way to study language diversity with a multi-disciplinary view" (L. Besacier, keynote speech at this conference). It aims (i) to provide an illustration of the way in which data are collected in fieldwork on endangered languages, bringing attention to the quality of the transcriptions and annotations created by linguists, (ii) to present the contents and format of a set of endangered-language documents synchronizing sound and text, which are currently available online, and (iii) to sketch out some of the research purposes and applications to which these documents lend themselves, and which we intend to pursue in future work.