ElevenLabs introduces Cue, a forced alignment model that aligns an hour of audio in under a second

Details

ElevenLabs launched Cue, a forced alignment model that places every character in an audio file on a precise timeline, powering the Scribe model family and the Forced Alignment API.
Cue aligns one hour of audio in less than a second on production GPUs, compared to nearly 2 seconds for the next fastest tool (Qwen3-ForcedAligner) and over a minute for the slowest tested.
On real-world recordings, Cue achieved an average timing error of 179 ms across 29 benchmark languages, outperforming open-source tools that ranged from 283 ms to over 6,000 ms.
The model handles recordings of any length in a single pass, avoiding the cumulative timing errors associated with chunking long files.
Cue supports over 90 benchmark languages with a single model, whereas most competitors require separate models per language.
In music alignment tests using the JamendoLyrics MultiLang dataset, Cue placed 92.7% of words within 0.3 seconds of the label in English, French, Spanish, and German, compared to 82.0% for the next best tool.
The team resolved a performance issue in Georgian and other low-resource languages within 24 hours of customer feedback by training and releasing a new model version.

Key point