How to Set Caption Start and End Times That Match Speech

Before touching a single timecode, make sure the transcript is finished. Timing a draft that still has gaps means redoing the work later, because every wording change shifts line lengths. I ask learners to read the text once against the audio, fix what they hear wrong, and only then open the subtitle editor.

Pick a test clip of about two minutes with a steady speaker. Time that clip completely before attempting a full recording. It shows you how your editor behaves, how fast the speaker really talks, and where your habits need adjusting. A short clip also gives you a fair measure of how long the full job will take.

Keep headphones on and the video visible. Captions are read while watching, so judge them against the picture, not against the waveform alone. The waveform helps you find the exact start of a word, but the picture tells you whether the viewer sees the text in a comfortable moment.

• Finish and proofread the transcript first
• Choose a two-minute test clip
• Use headphones and keep the video in view

A caption is usually one or two lines on screen at once. Split text at the points where a person would breathe or where the grammar allows it: after a comma, before a conjunction, between a subject phrase and its verb phrase only when you have no better option. Never leave a single dangling word like an article at the end of a line.

Aim for lines that look balanced. If the top line is much longer than the bottom one, the eye jumps awkwardly. A shorter top line with a slightly longer bottom line often reads more smoothly, though the rule is flexible. What matters is that the viewer never has to reread a line to understand it.

Keep a name, a number with its unit, or a verb with its particle on the same line. Splitting them forces the reader to hold half a thought across a line break. When a sentence is long, break it into two captions at a clause boundary and let the second one carry the continuation naturally.

• Split at pauses and clause boundaries
• Keep names and numbers with their units together
• Avoid ending a line on an article or preposition

The start time of a caption should land at, or a hair before, the first spoken word. Viewers notice text that arrives late far more than text that arrives a fraction early. Zoom into the waveform, find where the first syllable begins, and place the in-point there, then play it back to confirm it feels right.

Resist the urge to show text well ahead of speech. A caption that appears a full second before the words are spoken gives away what is coming and can mismatch with the speaker's expression. The goal is to feel simultaneous. A small lead of a few frames is acceptable and often feels more natural than exact alignment.

When a speaker starts after a pause, make sure the previous caption has cleared. Leaving old text hanging while new speech starts confuses the viewer about who is saying what. Clear the screen, then bring in the next caption cleanly with the new voice.

• Zoom into the waveform to find the first syllable
• Prefer a tiny early lead over a late arrival
• Clear the previous caption before a new speaker starts

The end time decides whether a caption can actually be read. A short caption needs a minimum display time even if the speaker finishes quickly. As a working habit, keep each caption on screen long enough to read it aloud once at a comfortable pace, and check the reading speed indicator your editor provides if it has one.

If a caption disappears the instant the last word ends, the viewer may not finish reading. Extending the out-point slightly past the speech, as long as it does not collide with the next caption, gives breathing room. Leave a small gap between consecutive captions so the change is visible and the eye registers new text.

Fast speakers are the hard case. When speech outruns comfortable reading, trim filler words that carry no meaning, such as repeated false starts, while keeping every word that changes the sense. Edit for clarity rather than transcribing every stumble, and follow the house style you agreed for the project.

• Give every caption enough time to be read once
• Leave a short gap between consecutive captions
• Trim filler only when it does not change meaning

Long silences deserve a decision. If the speaker pauses mid-sentence, you can hold the caption through a short pause, but for a longer one it is better to clear the text and bring back the remainder when speech resumes. Text left on screen over a long silence looks like an error.

Cuts in the picture matter as well. Where possible, avoid letting a caption straddle a hard cut, because the eye resets at the cut and the text can feel as if it is being read twice. Ending a caption just before a cut and starting the next just after it makes the viewing smoother, provided it still matches the speech.

When music or background noise carries on without speech, there is nothing to time, but check whether a sound note is needed. That is a separate decision from timing, and your editor should treat it as its own caption so that spoken lines stay uncluttered and easy to follow.

• Clear the screen over long silences
• Avoid captions that cross hard cuts where you can
• Keep sound notes as separate captions

When the timing is done, watch the whole clip at normal speed without stopping. Do not edit during this pass. Simply note every spot where a caption felt late, early, too fast to read, or awkwardly broken. Mark each with its timecode so you can fix them in one sitting afterwards.

Then play it once more with the sound off. If the captions alone tell the story clearly, your breaks and timing are doing their job. If you find yourself rereading, a line is probably too long or leaving the screen too soon. This silent pass catches problems that the ear forgives.

Finally, save your project and export a test file in the format you plan to deliver. Open it in a different player to confirm the timing survived the export. A file that looks right in the editor can drift in another program, and finding that before delivery saves an awkward correction later.

• Watch once at normal speed and only take notes
• Watch once with the sound off
• Test the exported file in a second player