Respeaking for the BBC
By Alison Marsh (Red Bee Media)
Based on the transcription of talk given in Forlì on 17th November 2006.
Keywords: viavoice, k-live, live respeaking, research, scripting, anaphoric presupposistion, training, multi-skilling, future, rispeakeraggio, trascrizione, assunzione, formación de traductores, formazione degli interpreti
©inTRAlinea & Alison Marsh (2006).
"Respeaking for the BBC"
inTRAlinea Special Issue: Respeaking
Edited by: Carlo Eugeni & Gabriele Mack
This article can be freely reproduced under Creative Commons License.
Permanent URL: http://www.intralinea.org/specials/article/1700
Subtitling by respeaking and voice recognition
Red Bee Media used to be part of BBC but now is a separate company. BBC is Red Bee Media’s biggest client and all the subtitling at BBC is made by Red Bee Media. First of all, let me explain how respeaking works at Red Bee Media. We use a voice recognition software which is called ViaVoice. Respeakers train the software to recognise their voices. They do a lot of training using ViaVoice so that they can improve their voice model to get it to on-air standards. They also have to use ViaVoice for a lot of preparation because ViaVoice has a very big dictionary, of about 75,000 words. Although it has this big dictionary, it is very English-focussed. But since Red Bee Media covers a lot of live sporting events, news broadcasts, because ViaVoice does not contain a lot of the vocabulary we need (e.g. footballers names or places in Iraq), respeakers have to do a lot of training into ViaVoice so that they can go on-air and subtitle these broadcasts.
Once they go on air they use a piece of software which is called K-live, especially designed for BBC’s respeakers by the BBC’s Research and Development department (when respeakers were part of the BBC). It allows respeakers to do whatever they want to do in terms of transmitting subtitles: positioning them on the screen appropriately; broadcasting them in different colours to indicate different speakers; making subtitles more accurate thanks to an automatic correction function; and it also allows respeakers to connect to any of the channels to be subtitled on. Red Bee Media subtitles on the two analogue BBC channels (BBC One and BBC Two) and digital and satellite channels like BBC News 24, BBC Parliament and BBC Three.
When a respeaker goes on-air s/he connects to K-live, put his/her headphones on and s/he respeaks what s/he hears into the microphone and s/he and then the voice recognition element of ViaVoice processes the vocal input from the microphone, turns it into text and K-live broadcasts it on the screen so that the words scroll out one by one. So, respeaking is essentially repeating what is heard, inserting the right punctuation, editing when necessary (condensing information to keep up with the speakers because sometimes they speak very fast especially on news broadcasts). In doing this, respeaking involves a complex mental process: hearing one thing, saying another thing, whilst keeping the main ideas and trying to keep a bit the flavour of the programme for the deaf viewers so that they can appreciate a little bit the way a speaker is speaking, the general register of the programme because a football commentator during a match speaks in a totally different way to a politician in the Houses of Parliament.
The history of live subtitling
As far as the background to respeaking, it came into being for different reasons. BBC has always been at the forefront in terms of subtitling. The first subtitling departments were set up in London and in Glasgow in the 1980s but initially they only did pre-recorded subtitling. There was no live subtitling. In 1990, the live subtitling unit was created, on a very modest scale. There was no respeaking since all the real-time subtitling was done by stenographers (former court reporters using phonetic keyboards in order to produce texts at a speed of about 250 wpm). Gradually, the live subtitling department began to expand and it started taking on more and more live output. In January 2001, development into respeaking began and this happened basically for three reasons: the first was that there was a growing demand for subtitles from the deaf community; the second was some legislation introduced by the Government called the Broadcasting Act, in 1990. This legislation stipulated that all of the major television companies had to increase the proportion of output that they subtitled (and this meant live subtitling as well as pre-recorded subtitling) up to 90% by 2010. Then the BBC set its own target, 100% on its analogue channels by 2008; thirdly, the BBC had to find an alternative way to cover all its live output because stenography is a very specialised skill, you have to train for at least five years to become a professional stenographer. As a result there were not many stenographers available, and the few could demand for very high salaries because it is a very specialised skill. Clearly, it was not possible for BBC to cover all its live output with stenographers because this was not financially viable. So, experiments began in the field of respeaking which was a more practical and cost effective way of subtitling large volumes of live output.
The first live respeaking was in April 2001 when the World Snooker Championships were subtitled. Later that year BBC subtitled the coverage of tennis, Wimbledon, which is a very important event in the English sporting calendar. So, BBC began subtitling sports. But gradually respeakers started subtitling more and more output including coverage of BBC Parliament, regional news and later, national news on BBC News 24. As they started knowing more about respeaking, the quality of the subtitles produced by respeaking increased. A very important factor for that was the software they were using, ViaVoice, the quality of which was improved by the introduction of ViaVoice 10, a new version. Then, with the introduction of K-live respeakers began having even more accurate subtitles.
As for research, we work in cooperation with the deaf community. We have done a lot of research in the readability of subtitles and we receive a lot of feedback by the deaf about the colours, the kind of subtitles they prefer (block or subtitles the scroll one word at a time). The result of this research has been the creation of house-styles. So we use punctuation in a given way, the number of words to be displayed on the screen to make sure that the deaf viewers can read subtitles easily. Another result we got from the deaf community at the beginning of our experimenting live subtitling was the option for immediacy over accuracy.
Currently, Red Bee Media subtitles through respeaking approximately 650 hours per month of live programmes. This means that in summer, when there is a lot of sports, respeakers subtitle more than that but in wintertime they subtitle less because there is less live output to cover. Generally, 24 hours a day of lived subtitling are covered by respeaking on a variety of channels (BBC one, BBC two, the digital and satellite channels). Moreover, since every individual British region has its five individual news per day, even the majority of regional news are subtitled through respeaking. At some times of the day there are 17 different regional programmes going on air simultaneously that would be subtitled by respeakers.
Respeaker training and professional aspects
Since then the amount of coverage increased and the kind of programmes has become more varied. By the time, as respeakers become more experienced and the quality of subtitles they produce increases, respeakers have started covering output the was traditionally the domain of stenographers and programmes such as the national news, current affairs programmes like “News Night” and “Question Time”. Then the number of respeakers has increased quite dramatically over the past three or four years. Now at Red Bee Media there are 50 respeakers working all over the country. The two main offices are in London and in Glasgow, but there are also respeakers working in small offices all around the UK. The software they use, K-live allows people to work even from home so that a number of respeakers can go on air from their living room.
Another important aspect of a respeaker’s job is something called scripting, contributing to the pre-recorded subtitling process. Voice recognition is used to produce the scripts of pre-recorded programmes. These scripts are then processed to create subtitle files which are then checked through by pre-recorded subtitlers. So respeakers are used in as many ways as possible. Leaders try to avoid respeakers going on air for too long during the course of the day because there is a lot of strain in the voice and every respeaker has to concentrate very hard to respeak live output. A respeaker is generally given half of live output and half of scripting every day.
As far as recruitment is concerned, in the past, respeakers were recruited from the ranks of subtitlers. The recruitment procedure was non-existing. Any subtitler with an interest in respeaking could have a try. However, now, the recruitment procedure is very stringent and respeakers are recruited externally most of the time. Candidates have to pass a series of tests before they can become respeakers. The most important test is called speech test. They are given samples of live output and they have to try respeaking it to see how they get on.
Similarly, in the past, the training was very ad hoc. There was not a structured training programme. It was not very clear how long it would take to each candidate to be able to go on air. Nowadays, the training programme is very organised and structured. Leaders know exactly what a respeaker is going to do during the course of their training which nowadays takes between two and three months. Their training is very closely monitored. All that has evolved as the department has expanded. There is also a reviewing system which is in place to ensure that respeakers are consistently producing high quality subtitles (the accuracy target is 97%, that is that over 100 words, three words can be incorrect).
Another important aspect is the so called multi-skilling, that is, instead of having separate file and live teams, Red Bee Media tries to combine the two so that every subtitler has experience of respeaking and every respeaker can do pre-recorded subtitling as well. This is to make our workforce as flexible and productive as possible, thus allowing Red Bee Media to attend the BBC’s 100% target by 2008.
In the future Red Bee Media live subtitling offices are continuing to expand, particularly those in Glasgow. They are taking more and more flexible people as possible and every person is trained on every aspect of the job. One of the advantages of having this flexible workforce is that respeakers have to work shift works because so as to cover nearly 24 hours on the BBC. This involves very early news broadcasts and very late sport broadcasts. So, since every respeaker has to work a variety of shifts including week ends, the larger the workforce that can cover different shifts, the easier it will be for individuals to work more dayshifts and not so many demanding shifts. Lastly, Red Bee Media is experimenting with respeaking in another language. A lot of respeaking has been made in French but they are looking into respeaking in Spanish and German.