Open archive of 240,000 hours' worth of talk radio, including 2.8 billion words of machine-transcription

A group of MIT Media Lab researchers have published Radiotalk, a massive corpus of talk radio audio with machine-generated transcriptions, with a total of 240,000 hours' worth of speech, marked up with machine-readable metadata.

The audio was scraped from streaming radio services between Oct 2018 and Mar 2019, and the transcripts run to 2.8 billion words. The researchers hope the corpus will be used by "researchers in the fields of natural language processing, conversational analysis, and the social sciences."

I'm mostly interested in the social science implications here: talk radio is incredibly important to the US political discourse, but because it is ephemeral and because recorded speech is hard to data-mine, we have very little quantitative analysis of this body of work.

As Gretchen McCulloch points out in her new book on internet-era language, Because Internet, research on human speech has historically relied on expensive human transcription, leading to very small and corpuses covering a very small fraction of human communication.

This corpus is part of a shift that allows social scientists, linguists and political scientists to study a massive core-sample of spoken language in our public discourse.

We introduce RadioTalk, a corpus of speech recognition transcripts sampled from talk radio broadcasts in the United States between October of 2018 and March of 2019. The corpus is intended for use by researchers in the fields of natural language processing, conversational analysis, and the social sciences. The corpus encompasses approximately 2.8 billion words of automatically transcribed speech from 284,000 hours of radio, together with metadata about the speech, such as geographical location, speaker turn boundaries, gender, and radio program information. In this paper we summarize why and how we prepared the corpus, give some descriptive statistics on stations, shows and speakers, and carry out a few high-level analyses.

RadioTalk: a large-scale corpus of talk radio transcripts [Doug Beeferman, William Brannon and Deb Roy/Arxiv]

(via Four Short Links)