Speaker 1: Hello, my name is Pablo Romero Fresco. I work at the University of Vigo in Spain, where I lead the research group Galma, and where I conduct research on media accessibility. In this first video lecture for Module 2a, we'll be introducing re-speaking as a form of live subtitling. We'll provide a definition, compare it to other live subtitling methods, we'll see a few examples of how it works and how it differs from country to country, and we'll go over how and where one can be trained as a re-speaker. Live subtitling may be defined as the real-time transcription of spoken words, sound effects, relevant musical cues, and other relevant audio information. Live subtitles can be produced through different methods, such as VeloType, which is a special keyboard used in countries such as Holland, and which allows for fast typing. DualKeyboard, which was used a few years ago in countries such as Germany, and which involved subtitlers taking turns to produce live subtitles. And there's also a Fiway keyboard system. This used to be quite common in Catalonia, now it's been replaced by speech recognition. This system involves five typists working together. They're listening to the audio when their screen shows red, they start typing when the screen goes green, and they stop again when they see the red colour, so that their colleague sitting next to them can take over. Stenography is a system that allows an operator, called a stenographer, to write phonetic shorthand at a special keyboard. Pressing multiple keys simultaneously, they manage to spell out whole syllables, words, or phrases with a single hand motion. And then we have re-speaking, which is what we will cover in the next slides. As part of this module in the ILSA course, you'll find different videos illustrating some of the live subtitling methods that we have just described. The first video here shows an event organized by Intersteno, an organization that brings together speech-to-text reporters using different methods. The second and the third videos here show how the intersteno system works. The second and the third videos here, they are material produced by the Australian and the Canadian governmental regulators to show users how their live TV programs are subtitled on a regular basis. They cover both stenography and re-speaking. And then we have a last video, that is one of the first videos ever made about re-speaking by the BBC. The sound quality is not too good because this is the only copy available, but it does help to get an idea of what re-speaking is and about the transition from stenography to re-speaking in some countries. Now, comparing the different live subtitling methods is useful to see what parameters are important in live subtitling and in live subtitling quality, including delay, speed, accuracy, difficulty, cost and errors. Later, we'll see that errors and delay are the key elements that determine how re-speaking is approached in every country. But for the purpose of this table, and especially comparing the most preferred live subtitling methods, which are stenography and re-speaking, I would like us to focus on both difficulty and cost, if possible. Now, while both stenography and re-speaking can be accurate, the difference between them lies in the difficulty and the cost. As I was saying, it takes about between three to four years to train a stenographer who can demand a high salary, while a re-speaker may be trained in two to six months, and often they have lower salaries. As a result of this, and also as a result of the constant improvement of speech recognition technology, re-speaking has overtaken stenography as the preferred form of live subtitling, although it really depends on the country. In North America, that is in the US and Canada, and in Italy, stenography, at the moment, is still the preferred method, although we're beginning to see a transition towards re-speaking. Australia and the UK use a combination of both methods, although, for instance, in the UK, 10 years ago, say in 2010, there was around 25% re-speaking and 75% stenography. Now, the distribution is the opposite, so tables have turned completely. In Spain, Switzerland, Belgium, France, most of these countries resort almost exclusively to re-speakers, at least for TV subtitles, and stenography is still the preferred method to make live events accessible in these countries. Now, re-speaking may be defined as a technique in which a re-speaker listens to the original sound of a live program or a live event and re-speaks it, including punctuation marks and some specific features for the deaf and hard-of-hearing audience, to a speech recognition software, which then turns the recognized utterances into subtitles, displayed on the screen with the shortest possible delay. About the origins of re-speaking, well, court reporter Horace Webb explains that re-speaking was born in a Chicago courtroom in the 1940s. He says its father was a pen shorthand reporter and its mother frustration. Till then, court reporting was done manually, and then it was then when Horace Webb started re-speaking the audio of the court sessions on a tape, a technique that was called voice writing. There was no speech recognition involved there. The first person to speak to use speech recognition for this purpose was Chris Ailes in 1999. Chris Ailes is now a voice writing trainer, a real-time voice writing trainer, and some of the materials that she uses are actually included in the ILSA course. Now, Chris Ailes, as I was saying, in 1999, she re-spoke the audio of the court sessions to the speech recognition software Dragon NaturallySpeaking. This is a technique that is known in the U.S. as real-time voice writing. In Europe, the first re-spoken TV subtitles were produced in 2001 by the BBC in the UK, and in the same year by Wiarté in Flanders in Belgium. Wiarté is one of the partners in the ILSA project. In this video, you can see how re-speaking is produced, and it's actually a video by the BBC where it shows how Martin Conwell, a re-speaker, goes about his daily job.

Speaker 2: Also, we have invested in new road projects in and around the area, which have been called for for years. Most people think subtitlers do an awful lot of typing, whereas actually we produce subtitles using voice recognition software and a technique that we call re-speaking, where I will have an audio feed of what's coming on through the TV, and I will speak along with that input, repeating it. We need to make sure that whatever changes we make have children PASMAC rights at the heart. It doesn't sound much like normal spoken English. We have to speak in this monotone way, giving equal weight to each syllable we speak. So Conwell, you know how the game works, full stop. Let's meet the usual suspects, full stop. This is because the computer can't understand a lot of the intonation of normal speech. Today I'm starting off doing a piece of pre-recorded television. It's a BBC...

Speaker 1: You'll find in the section devoted to this unit, and to the resources included in this unit, in the Ilse platform, different videos showing how re-speaking is done in different countries, such as the UK, the US, France, Canada, Italy, Poland. In the UK and the US, re-speakers speak into the microphone and if there are any errors, they attempt to correct some of them once they have been displayed for all the viewers. Now in contrast, in France, between three and four people are involved in the production of live subtitles. Once the re-speaker has dictated to the speech recognition software, the subtitles are read by a second operator called the whisperer, who whispers potential corrections to a third operator called the corrector. This is a way to ensure high accuracy in the production of live subtitles, although understandably this comes at the expense of increasing the delay of the live subtitles. In the UK, where self-correction is used, the average delay of live subtitles ranges between three to six seconds. In France, live subtitles can sometimes be delayed by more than 10 seconds. The video produced by a company in the French-speaking part of Canada shows how this broadcaster, VAT, uses a Nintendo 64 joystick to introduce punctuation, mostly full stops and commas in the subtitles. Punctuation marks can sometimes be recognized as words as opposed to commands, so this is a way to ensure that your errors are minimized or fully eliminated. Finally, one of the videos that we have is a video produced by an Italian company called Kolbe, and in this case what they use is a touch screen to allow re-speakers to introduce proper nouns, specialized names, colors for the speakers, etc., ensuring that no errors are caused by misrecognition. Needless to say, this system adds a layer of complexity to what is already quite a challenging task. Re-speakers must combine, in this case, dictation with the manual selection of keywords. In any case, these videos illustrate that the way in which every country or even a company tackles the issue of accuracy and delay or the issues of accuracy and delay very much determines the particular speaking, re-speaking approach that is adopted. So it is the handling of accuracy and delay that really makes a difference in terms of how one goes about re-speaking. As far as the professional practice of re-speaking is concerned, recent data obtained within this project, the ILSA project, shows that re-speakers are normally highly educated women under 40 with a BA and language-related MA. They work as re-speakers for around six years and normally for subtitling companies, whether in-house or freelance. There is commonly, there is a demand for quality, but this has changed quite a lot because now re-speakers often work with scripts that help them a lot. There are key differences in equipment depending on the country, sometimes depending on the company. Accuracy and delay are the key issues and they are approached differently from country to country. And re-speakers, we know they don't normally re-speak for longer than, say, 40 minutes in an hour. If the shifts are longer, then two re-speakers are needed. That is the recommendation. As far as training is concerned, we could argue that re-speaking is to subtitling what interpreting is to translation. That is, it's a leap from what's written to what's oral, but not having the safety net provided by time. Now, in other words, the process involved in re-speaking is a process of interpreting, or more accurately, simultaneous interpreting. However, the result of this process is the production of subtitles and the method involves the use of speech recognition software. It follows then that training and re-speaking may need to involve elements of interpreting, subtitling, and some re-speaking specific aspects related to the use of speech recognition software. Now, research skills, split attention, and pressure management are important competences from interpreting. They may be very relevant to re-speaking. Then we have reformulation, spelling, punctuation, pressure management, subtitling for the deaf and hard of hearing, issues about that, such as speaker identification and subtitle positioning. So those are all very, very important. And then from re-speaking, there is the very important skill related to the use of the software. Then we need fast and accurate dictation, and of course, we need the ability to be able to multitask as you are re-speaking. We will cover that in some of the coming units in this module. Now, re-speaking training at university is limited, perhaps with the exception of a few re-speaking modules that are normally included within larger MA programs, such as audiovisual translation programs or interpreting programs. There is introductory training at the University of Leeds in the UK, University of Barcelona, the autonomous University of Barcelona, and in Parma in Italy. There is more thorough training in Belgium at the University of Antwerp, which is one of the partners in this project. The same goes for the University of Warsaw in Poland, six months each. And we have the University of Roehampton in London, where the training lasts for three months. And then there is the University of Vigo, which is offering the first online re-speaking course, which is actually an interlingual re-speaking course with an interlingual component and is based on the material produced here in the ILSA project and lasts for around five months. This is all for this first video lecture in this module 2a interlingual re-speaking. There are other video lectures within this module and if you want to have a look at them, then feel free to have a look at our units and check out the video lectures. Thank you very much for your attention.