ElevenLabs Cue forced alignment: better subtitle timing starts with the right text

ElevenLabs’ October 28, 2026 research update describes Cue, a forced-alignment model that places transcript characters on an audio timeline. The company says Cue supports its Scribe transcription family and is also available through its Forced Alignment API.

For video creators, the useful distinction is between knowing when words are spoken and deciding how those words should appear on screen. Alignment can help establish timing; readable captions still require attention to the text, the picture and the audience.

Separate transcription from timing

Transcription supplies the words. Forced alignment connects a supplied transcript to the recording. Begin by correcting names, missing words and punctuation so the text actually corresponds to the performance you intend to publish.

If a sentence is wrong, making its timestamps more precise does not fix its meaning. Listen to uncertain passages rather than treating a clean-looking transcript as proof of accuracy. Keep the corrected text with the specific audio file it describes.

Align the final audio version

Use the version of the voiceover or dialogue that will appear in the edit. Removing an opening pause, replacing a line or changing playback speed afterward can move the relationship between words and timestamps.

A useful handoff includes the audio version, matching transcript, alignment output and caption export. When the recording changes, review whether those timing files still describe it. This small piece of version discipline avoids carrying yesterday’s offsets into today’s cut.

Try a three-phrase timing review

An original practice script is: “Wait here. Listen to the rain. Now open the door.” Record or select a performance with deliberate pauses, then review the beginning and end of each phrase against the audio.

Watch once while listening and once while concentrating on the captions. Does a phrase arrive early enough to reveal the next action before it is spoken? Does it remain after the shot has moved on? This is a suggested editing exercise, not a tested result from Cue.