Our Reader in Culture, Media and Creative Industries Dr Richard Howells continues his work with the British Library (BL)on archiving speech and moving pictures –and especially speech to text transcription technology for scholarly research. This month he attended both the BL’s advisory board on the project, and an invitation-only conference “Opening up Speech Archives.”
The BL already has over a million speech-based recordings in its archives, together with more than 25,000 hours of television news. The challenge now is how to make this all searchable as an on-line resource.
The main problem here is that current speech-to-text transcription software is not very accurate (between 60 and 90 per-cent- though actually probably nearer the lower limit). As Oxford University’s Peter Robinson told the conference, while perfect-speech-to text transcription remains “the Holy Grail”, there is “still no magic bullet.” Prominent among other contributors were the BBC’s Research and Development Unit and the Netherlands Institute for Sound and Vision.
Despite the technical challenges, the project is determined to make the archive as useful as possible for researchers now, even while the technology is still improving. As the BL’s Lead Curator of the Moving Image Luke McKernan told the conference, sound and vision need to be placed on an equal footing with traditional library materials in terms of our ability to search them. There was “a great change” afoot, he said, and speech-to-text technology was at the heart of this.
Richard Howells’ part in this project is not as a technologist but as an academic user who continues to champion the use of audio and visual material for scholarly research, especially in the humanities.
The project is funded by the Arts and Humanities Research Council. Howells has been a member of the project’s advisory board since 2012.