Each line of the transcript must sit on the media clock
Aliases: interactive transcript · time-stamped transcript · forced alignment
What it is
If the transcript and the sound are merely “two versions of the same session,” people cannot meet between them. Time alignment (a time-aligned / interactive transcript) sits every sentence on the media clock: click the line, the playhead lands on that second; the line that is playing is marked. A wall of text satisfies “there is a transcript.” It does not satisfy “it maps.”
Why it happens
Media is time. A document is space. Without alignment, someone who has found a passage still scrubs by guesswork; error is often tens of seconds to minutes. Once aligned, the transcript is also navigation: listeners locate with the eye, readers can sit on a sentence without treating the player as a cassette.
Granularity sets the use. A stamp per minute is better than none and only coarse. Sentence-level hits the proper name. When the text already exists, forced alignment maps words onto the waveform at far lower cost than recaptioning. The characteristic failure is a whole-block shift — click “conclusion,” land on the opening small talk. Users blame the wording. The clock was wrong.
Studying it
Set find-and-seek tasks: click the target sentence in the transcript, measure error between the playhead and the true onset. Compare no stamps, paragraph stamps, sentence stamps.
Independent variables: alignment grain, whether the current sentence highlights during play, automatic alignment versus hand-spotting. Dependent variables: seek error in seconds, number of actions, clicks on the wrong sentence.
Do not score “can you find this word.” That is search, not alignment. The distribution of error beats the mean: a few large shifts destroy trust.
Where it stops holding
Minutes that will never be consumed with the audio care more about date and attendance than timestamps. A live session whose clock is still running cannot hold a stable alignment. Music, applause and piled-up voices blur sentence boundaries; sentence-level stamps then fake precision. When the user treats the transcript as an article and never opens the player, current-line highlight does nothing — the stamps still help a later listen. Perfect alignment of wrong words does not rescue content; correction is a different pipeline.
Applying it
- Put clickable timestamps on sentences or paragraphs. Highlight the current sentence during playback.
- After automatic alignment, spot-check opening, middle and end. A whole-block shift shows up in one pass.
- How to check: click a proper name that occurs once. The player should land within ±2 seconds of the real utterance. Play thirty seconds and watch the highlight follow. If it does not, the transcript is still a document sitting beside the media.