Skip to content

ElevenLabs releases Cue forced alignment model and API

ElevenLabs

The system aligns an hour of audio in approximately 0.5 seconds on a GPU across over 90 languages without chunking recordings.

Key points

  • Average timing error on full recordings is 179 ms, compared to 283 ms to over 6,000 ms for open-source tools
  • Aligned 92.7% of words within 0.3 seconds on a subset of the JamendoLyrics MultiLang music dataset
  • Provides character-level timestamp placement in a single pass for audio of any length
  • Available standalone via the Forced Alignment API and powers ElevenLabs Scribe speech-to-text transcriptions

Why it matters

Developers building transcription, dubbing, and subtitling tools can generate character-level timestamps across multilingual audio without splitting files into chunks.

Sources