Embedded Files
Contact
Contact
Daniel Hartwig (University Archivist, Special Collections & University Archives)
Team:
Team:
Daniel Hartwig (University Archivist, Special Collections & University Archives)
Josh Schneider (Assistant University Archivist, Special Collections & University Archives)
Jenny Johnson (Collection Management & Processing Archivist, Special Collections & University Archives)
Background
Background
This project will focus on using speech-to-text recognition to generate transcripts for 220 digitized audio recordings associated with the KZSU Project South Interviews.
During the summer of 1965, eight students from Stanford University spent ten weeks in the southern states tape-recording information on the civil rights movement. Sponsored by KZSU, Stanford's student radio station, the interviewers visited over fifty civil rights projects in six states and secured three hundred and thirty hours of audio recordings, including over two hundred hours of personal interviews.
Original recordings are housed in the Archive of Recorded Sound. Original transcripts are housed in the Stanford University Archives. Digitized audio and associated transcripts have been accessioned to the Stanford Digital Repository (SDR), and are currently made available via Searchworks/PURL and a Spotlight at Stanford exhibit.
Research
Research
The transcription of these digitized audio recordings will make them discoverable and analyzable by Stanford faculty, students, and other scholars.
Data
Data
Over 220+ hours of digitized audio recordings. 220 individual files.
Many of these recordings have multiple voices and background noise.
Desired Output
Desired Output
Accurate transcription of the audio recordings, which can be accessioned to the SDR, added to the existing Spotlight at Stanford exhibit, and subjected to additional analysis by scholars.
Experiments
Experiments
1) We will run the pre-built video model of Google Cloud Speech-to-Text recognition on 220 .wav files. (The video model is recommended for audio recordings that include multiple speakers).
Report abuse