What live subtitling is
Live subtitling is text produced while someone is speaking, with no chance to prepare or edit it first. News broadcasts, sports, debates, parliamentary sessions, lectures and many events are live, so the words are not known in advance. Prepared subtitles for recorded programs can be timed and checked carefully, but live text has to be produced on the spot.
The terms are used differently in different places. In some countries, “subtitles” and “captions” are close to interchangeable, while elsewhere captions mean text for deaf and hard of hearing viewers specifically, including sound descriptions. The article on captions versus subtitles sorts out the wording.
Why it matters
Deaf and hard of hearing people vary a great deal. Some use sign language as a first language, some rely on lipreading, some use hearing aids or implants, and many combine approaches. For many of them, text is a key way to follow live speech. Without it, a live program may be largely inaccessible at the moment the information matters, such as a weather warning, a public announcement or a political address.
The benefits include:
- Timely access. Viewers can follow events as they happen rather than waiting for a recording.
- Independence. A viewer does not need a friend or relative to relay what was said.
- Shared experience. People can take part in the same moment as everyone else, such as a match or a debate.
- Support for varied needs. Text can also help people who are not deaf, such as those watching in a noisy place.
Text is not the right access for everyone, and some viewers prefer a sign language interpreter. Providing choices is generally better than assuming one format.
How live subtitles are produced
Several approaches exist, each with different strengths. The broader methods are compared in the guide to live caption methods, and the basics are covered in live captioning.
- Stenography. A trained operator uses a specialized keyboard to type sounds in chords, and software turns them into words. The method is described under stenographic captioning.
- Respeaking. A trained person repeats what is said, adding punctuation, into speech recognition software tuned to their voice. See the respeaking method.
- Fully automatic. Speech recognition software produces text directly from the audio with no person in the loop.
- Hybrid approaches. A person may monitor or correct automatic output as it appears.
Which approach suits an event depends on its subject, the speakers, the audio quality, the budget and how serious an error would be. No method is perfect.
Delay
Live text is always behind the speech. The person or software needs time to hear, process and produce the words, and the pipeline that delivers the text to the screen adds more. The delay may be a few seconds and can be longer in difficult conditions. In a broadcast, the program may be delayed slightly to make up for it, but this is not always possible.
Delay matters because the text can arrive after a joke has landed, after a speaker has changed or after the picture has moved on. Viewers who are following both the speaker and the text may notice the mismatch most in fast conversation. The topic is discussed in depth under caption delay.
Some delay is simply a trade-off. Showing text sooner can mean showing words before they have been checked, which leads to more mistakes. Showing words in bigger, more complete chunks can be easier to read but arrives later.
Errors
Live subtitles can contain mistakes: misheard words, wrong names, missing words and odd punctuation. Common causes are overlapping speech, fast delivery, poor audio, accents, unfamiliar names and technical terms. Errors in live text are expected and are different from the more carefully checked work done for recorded material.
Some errors are minor and easy to understand in context. Others change the meaning, and a viewer may not know which kind they are looking at. For anything where accuracy matters, such as medical, legal or safety information, a qualified human should be involved, and corrections should be made afterwards where possible, as covered in post-event correction.
Helpful steps include sharing names and terms with the captioner beforehand and asking speakers to talk at a steady pace, which is discussed in the guide to speaker preparation.
What viewers can reasonably expect
Viewers can reasonably expect live text to give the gist and most of the detail, and they should not expect a perfect transcript. A realistic view helps everyone:
- Text usually trails the speech by a short time.
- Names, numbers and specialist words are the most likely errors.
- Quality varies with audio and method.
- Sound information such as laughter, music or an interruption may be missing or brief; see non-speech information.
Practical points for organizers
If you run a live program, plan for access early. Provide a clean audio feed, make sure the text is displayed in a readable size and position, share any names and materials in advance, and tell viewers how to report problems. Requirements for live access differ from place to place, so check with a qualified professional or the relevant authority about what applies to you.
Live subtitling is not flawless, but when it is well planned it lets deaf and hard of hearing viewers take part in events that would otherwise be out of reach.