Speaker of the House of Keys explains how voice recognition technology is being deployed in the production of the Official Report in the Isle of Man.
Mr Rodan is the Speaker of the House of Keys, the manx lower House.A Pharmacist, he was first elected to the Keys in 1995 and elected Speaker in 2006.
When it was suggested that speech recognition technology might be able to produce “instant” text for the offi- cial report of Tynwald,1
the idea
seemed as far-fetched as something out of science fiction novel. However, after two and a half years of hard work, science fiction became reality in April 2008, when, thanks to the vision and perseverance of senior Hansard editor Mr Ian Faulds, the Isle of Man became the first Parliament in the Commonwealth to enter the instant transcription age. However, as you might expect, it is a little more complex than that! In simple terms Members’ speech is dig- itally recorded and “converted”, through individual voice profiles, into continuously scrolling text.This is immediately available to the Hansard editors at a remote location, so they can begin the process of “tidying up” what Members have said within three or four minutes of the beginning of a
accuracy for automatic transcription of well spoken English, it has done rather less well for heavy accents, slurred speech, sore throats or lisps. Remarkably, it has coped well with Manx Gaelic when the relevant words have been given English phonetics. The fact that the resulting automatic transcription is not 100 per cent accu- rate means that all speeches do need to be edited before being published! Even a small error, such as omitting the word ‘not’, could be of major impor- tance when reviewing a debate. This has led to some light-hearted
parliamentary session.
The immediate benefit of the sys- tem has been twofold: to increase the speed of Hansard delivery to Members and the wider public, at the same time reducing the cost of pro- duction. For those of us interested in connecting Parliament to people these are important considerations.
The downside There is, however, a caveat.While the use of our speech recognition software has so far delivered up to 95 per cent
322 The Parliamentarian 2008/Issue Four
moments, particularly during the early stages of development, when it quickly became apparent that editors needed to remain alert to what was really being said. Did, for example, the Member really say “a pig of a floating pouch penalty”? Not quite. Clicking on the audio button in fact revealed they had said:“they pick up a £40 penalty”! The software,2
originally devel-
oped in Australia, relies on the use of individual voice profiles and these are “harvested” – to use a technical expression – by asking each Member to record a five-minute pre-prepared