Why speakers matter so much
The people who talk on stage or on a call have more influence on live caption quality than most equipment. A trained captioner or a recognition tool can only work with what it hears and how quickly it arrives. Fortunately, the habits that help are small, easy to explain and mostly the same as those that make a talk easier for the whole audience.
None of this requires speakers to sound unnatural. The aim is clear, steady, well-prepared delivery, not a robotic one.
Habits during the talk
Keep a steady pace
Speed is the biggest challenge for any live method. A fast speaker creates more words to catch and increases caption delay. Very slow, exaggerated speech is not needed, and can actually confuse recognition tools, which are trained on natural speech. A comfortable, conversational pace with short pauses between ideas is ideal.
Pause at natural breaks
A pause at the end of a thought lets a human writer finish and gives recognition tools a chance to settle on punctuation and sentence boundaries. It also gives the audience time to read. Reading speed matters for live text, and it is easy to overload viewers when a speaker rushes.
Speak one at a time
Interruptions and overlapping talk are among the hardest things to caption. In a panel, a moderator can help by inviting people in turn. When there is a lively exchange, it can be worth asking participants to wait until the previous person has finished.
Use the microphone properly
Hold or position the microphone consistently, close enough to be clear. Turning away, moving far from it, or covering it will drop the sound level and increase errors. If the venue provides a handheld microphone for audience questions, remind people to use it.
Repeat audience questions
A speaker who repeats a question before answering helps everyone: people in the room who did not hear it, people watching a stream, and the captioner who may not have received the audience microphone feed. It also makes the later transcript easier to follow.
Say who is speaking
In discussions, stating a name before speaking can help the captioner, since distinguishing voices is hard. Automated tools offer speaker labels but they can be unreliable, so naming yourself is a simple safeguard.
Spell out the unusual
If a name, acronym or technical word is likely to be unfamiliar, say it clearly and, where appropriate, spell it once. Place it on a slide if possible. This helps human captioners and automatic tools equally.
Mention what is on screen
When pointing at a slide, chart or object, describe it in words. Someone reading captions may not be looking at the screen, and a description helps people who are blind or have low vision too. It is a small step toward the habits behind audio description.
Be careful with sounds and jokes
Laughter, applause and asides are easy to lose in live text. If a sound matters to the meaning of a moment, say so. Wordplay and quick asides tend to survive poorly in live text, so explaining a joke briefly may be worthwhile for a captioned audience.
Preparation before the event
Send materials early
Captioners and technology teams benefit from the following, as early as possible:
- slides or a script
- a list of names, acronyms and technical terms
- the running order
- pronunciation notes for unusual names
Human writers can load terms into their dictionaries ahead of time, and some automatic tools accept a custom vocabulary. A shared term glossary makes all these steps quicker. Late changes should be passed on too.
Share a script with care
If a speaker reads from a script, sending it helps a great deal, but speakers often depart from the text, so the captioner should be told that it may not be followed exactly. Reading aloud from a script also tends to speed speakers up, so a reminder about pace is useful.
Test the setup
If possible, have speakers do a brief sound check with the actual microphone and some of their own material. Hearing a few seconds of the live text can show them where they are speeding up or mumbling.
Plan for videos and audio clips
Pre-recorded videos shown during a talk should be captioned in advance. If a clip is not captioned, a live captioner may be unable to cover it, especially if its sound is played into the room and not into the captioning feed. Check that the audio from clips reaches the captioner.
Speakers who are remote
For online speakers, a headset microphone in a quiet room is a better choice than a laptop microphone. Closing other applications, muting notifications and using a wired connection where possible all reduce problems. Speakers should know in advance who will operate the captions, and what to do if they stop.
Things to avoid asking
Speakers should not be asked to change their accent, speak unnaturally or avoid the vocabulary of their field. Some voices are harder for software, but the burden should fall on the organizers to choose appropriate methods and give preparation material, not on the speaker to sound different. If a speaker has a speech difference or uses a language that automatic tools handle poorly, a human captioner is often the better option, and the speaker should be asked what works for them.
A one-paragraph briefing
A short note to speakers might read: speak at a comfortable pace, pause between ideas, use the microphone, repeat questions from the audience, say your name when you start, spell unusual terms, and send us your slides and a list of names and terms in advance. That is usually all that is needed to make a visible difference.