ElevenLabs Cue Maps Every Character to the Audio Timeline.

Cue is ElevenLabs’ new forced alignment model for placing every character from a transcript on an audio timeline. It handles recordings of any length in one pass and already powers every Speech to Text transcription on the platform.

Forced alignment connects written transcripts to the points where words were spoken. Those timestamps keep subtitles on time, match dubbed lines to speakers and let someone click a word in a transcript to jump to that moment in the recording.

In its LinkedIn post, ElevenLabs reports that Cue placed characters more precisely than every open-source aligner it tested.

Most open-source alignment models tested by ElevenLabs were built around short, clean audio in a limited set of languages. Cue uses one model for recordings of any length rather than requiring long files to be divided into shorter clips before alignment.

The research benchmark covered real-world audio in 29 languages. Cue was more accurate than every open-source aligner tested across all 29. On one GPU, the model can align an hour of audio in about half a second.

Benchmark languages

29

Cue led the tested open-source aligners in each language

Audio processed

1 hour

completed in about half a second

Hardware

1 GPU

for the reported processing time

The output works at character level. Each character is assigned a position on the audio timeline, giving applications the timing data needed to place complete words correctly.

Cue is not limited to ElevenLabs’ full Speech to Text workflow. Builders can access the model separately through the Forced Alignment API when they already have a transcript and audio file but need timestamps connecting the two.

That covers several workflows described in the announcement. A subtitle system can use the timestamps to decide when words appear. A dubbing workflow can match translated lines with the original speaker’s timing. An interactive transcript can connect each word with the corresponding point in a recording.

Because Cue already runs behind every ElevenLabs Speech to Text transcription, existing transcription output is using the new alignment model without requiring a separate alignment step. The standalone API provides the same alignment function as its own service.

What is ElevenLabs Cue?

Cue is a forced alignment model that maps every character in a transcript to its position on an audio timeline. This produces timestamps used for subtitles, dubbing and clickable transcripts.

Which languages does Cue support?

ElevenLabs benchmarked Cue on real-world audio across 29 languages. It reported higher accuracy than every open-source aligner tested in all 29 languages.

Is ElevenLabs Cue free?

The LinkedIn post does not include pricing or plan details. Cue is available separately through the Forced Alignment API and already powers every ElevenLabs Speech to Text transcription.