What delay looks like
Anyone who has watched a live news bulletin with captions on has seen it. A presenter finishes a sentence, the picture moves on to the next item, and the text for that sentence is still scrolling in. Sometimes it is a moment behind; sometimes the lag is long enough that the words no longer match what is on screen.
This is normal for live captioning, not a sign that something is broken. It happens because converting speech to text takes time at several points along the chain.
Where the time goes
Delay is the sum of several small steps. Different setups add different steps, but the usual ones are as follows.
Hearing and understanding
A person or a program has to hear enough of what was said to be able to write it. A captioner often waits for a phrase to become clear before committing to it, because a word spoken in isolation can be ambiguous. Automatic systems face the same problem: more context improves the guess, but waiting for context adds lag.
Producing the text
- A stenographer writes strokes that software converts into words. Fast writers can keep close to speech, but a short lag is still natural.
- In respeaking, the person must hear the speech, then repeat it, and only then does the recognition software convert it to text. That extra repetition adds time.
- Automatic recognition processes the audio in short chunks. It may show a rough result quickly and then revise it, or wait for a more stable result before displaying anything. Systems tend to trade speed against stability.
Transmission
If the captioner works remotely, audio travels to them over a network and text travels back. Each leg adds a small delay, and congested connections can add more. Encoding and delivering the signal to viewers adds still more.
Caption encoding and display
In broadcasting, text is inserted into the signal in a separate step, and viewers’ equipment may take time to decode and show it. Some displays also hold text briefly so that it can be shown in complete lines or blocks.
Display style
Captions can appear in roll-up style, where words scroll as they arrive, or in pop-on blocks, where a whole phrase appears at once. Pop-on can look neater but requires a complete phrase to be ready, so it is generally later than roll-up for the same speech.
Why the delay matters
Delay is more than an inconvenience. Several effects follow:
- Mismatch with the picture. When the camera cuts to a different speaker or shows a graphic, the text may still be explaining the previous moment.
- Missed context. Jokes, reactions and applause happen in the moment, and a late caption makes them harder to follow.
- Reading pressure. If text falls far behind, viewers may skip ahead or lose track, particularly when reading speed is already a limit.
- Lip reading. Viewers who combine captions with watching the speaker’s face notice mismatches more.
Pre-recorded content is different
Some programs that look live are in fact recorded shortly in advance or have a short broadcast delay built in. When there is time, captions can be prepared and timed to the picture, so they do not lag in the same way. That is one reason a scripted segment often captions better than an unscripted interview in the same bulletin.
Live captions can also be adjusted for the broadcast delay if the picture is deliberately held back for a short time by the broadcaster, though whether this happens depends on the station and the equipment.
What can reduce delay
Delay cannot be eliminated, but some choices help.
Clean, direct audio. A good feed straight from the microphone mix gives the captioner or software less to untangle, which reduces hesitation and corrections.
Preparation. Sharing names, topics and scripts in advance lets a captioner enter terms ahead of time, and lets automatic tools use custom vocabulary rather than guessing.
Steady pacing. Presenters who avoid bursts of very fast speech, and who pause between topics, give captioners a chance to catch up. A speaker preparing for live captioning can do more for delay than any equipment.
Choice of method. Different methods have different typical lag, and which is shorter in practice depends on the people, the tools and the connection. Testing before the event is more informative than assuming.
Reliable connections. For remote captioners, wired connections and a backup route reduce the chance of sudden extra lag or drops.
Display choices. Showing words as they are produced can feel faster than waiting for complete blocks, at the cost of some tidiness. Which suits the program depends on the audience.
Setting expectations
Producers can help viewers by being honest. A short note that captions are live and may lag or contain errors is better than leaving people to wonder. For important announcements, repeating key details in on-screen graphics, such as phone numbers, names and dates, means viewers do not depend on the caption alone.
Where captioning is a legal requirement, rules about acceptable delay and accuracy for live programs may exist and differ between places. The relevant regulator or a qualified professional is the right source, not a general article.
Delay and errors are linked
Faster output usually means less time to check. A captioner or program that chooses speed may produce more mistakes; one that waits for certainty may be later. Live captioning always sits somewhere on that line. Understanding that trade-off helps viewers and producers judge captions fairly: a short lag with mostly correct text is usually a good result for unscripted live speech.