The Smithsonian has hundreds of the earliest audio recordings ever made, but they have been considered unplayable and nobody knew what was on them. Recently, the Library of Congress and Lawrence Berkeley National Laboratory collaborated to make optical scanners capable of recording the patterns, and some of the results are becoming available. You can get a basic introduction to the situation at the National Museum of American History’s blog: Trilled R’s and the dawn of recorded sound in America, and Forgotten early sound recordings given a voice, and here‘s a YouTube playlist of six Volta Labs recordings; it sends shivers up my spine to hear a voice saying “It’s the eleventh day of March, eighteen hundred and eighty-five” (though the silly-sounding high-pitched trills somewhat ruin the spell). There are many more links at this MetaFilter post, where I learned about the restoration.
Why “IN AMERICA”? Was there an earlier one elsewhere?
I wish that NMAH blog had really emphasized that the trilled /r/s were used as a sort of diagnostic thing (the “check one, check two” of its day); I think there’s a nonsmall chance this will go viral as “Recordings show American English had trilled Rs in the nineteenth century.”
Dearieme: Yes, indeed. Firstsounds.org has recordings from France dated to 1857. Listening on crude laptop speakers, I find the second recording on the page to be the clearest: the tune is evident, even if the words are hard to understand, even given that I already know what they are. The Volta Laboratories sounds being linked to are from 1881.