ElevenLabs released Cue, a forced alignment model. It places every character in audio on a precise timeline. It works on any recording length in one pass. It aligns an hour of audio in under a second. It supports over 90 languages. It outperformed all tested open-source tools on accuracy. It handles noisy audio and overlapping speech better than competitors. It aligned songs with 92.7% word accuracy within 0.3 seconds. It is available via the Forced Alignment API. It already powers ElevenLabs’ Speech to Text transcriptions.

• 179 ms average timing error on full recordings
• 57 ms average error on noisy English audio
• 92.7% of song words within 0.3s of label
• <0.5s to align 1 hour of audio