Forced Alignment by ElevenLabs
Align speech audio to a matching transcript and get word and character timestamps plus loss scores. Built for tight subtitle timing, karaoke and interactive transcripts.
## Overview
Forced Alignment by ElevenLabs matches a known transcript to a speech recording. It places each word and character on a timeline and returns exact start and end times in seconds, plus alignment loss scores. This solves subtitle drift and makes click-to-seek transcripts work well.
## What you can build
- Word-accurate subtitles for video editors and streaming players
- Karaoke-style word highlighting from a lyric sheet and a vocal track
- Audiobook text syncing for page, sentence, or word highlighting
- Interactive transcripts where clicking any word seeks the audio
- Bleep or mute windows by timestamping sensitive words
- Precise clip cutting by selecting words, then trimming by their timestamps
## Inputs
- An audio recording to align.
- The transcript as plain text that matches the recording exactly as spoken.
- Don't include speaker labels like "Speaker 1:". Diarization is not supported.
## Outputs
- A JSON file with:
- A word list, each with the word, start time, end time, and a loss score.
- A character list, each with the character, start time, and end time.
- An overall loss score for the full transcript.
- All times are seconds from the start of the recording.
## Limitations
- The transcript must match what was actually said. Missing words, paraphrases, or reordered phrases reduce alignment quality.
- Diarization is not supported. Multi-speaker transcripts with speaker labels can produce unexpected results.
- Very noisy audio, strong music under speech, heavy reverb, or overlapping speakers can raise loss scores and shift boundaries.
- Clipped audio, long silences, and poor mic placement increase the risk of errors.
## Safety & compliance
- Only upload audio you own or have permission to process.
- Get consent from speakers before uploading recordings that include personal data.
- Don't use forced alignment as an identity or speaker-verification tool. It is built for timing, not biometric recognition.
- Review outputs before publishing. Timestamp errors can mis-caption speakers or mis-time sensitive segments.
Example prompts
Great starting points for forced alignment.
API quick start
Run forced alignment with a single API call.
{
"prompt": "Good evening. Tomorrow along the coast, e…",
"inputAudio": "https://your-cdn.com/input.mp3"
}curl -X POST "https://api.wiro.ai/v1/Run/elevenlabs/forced-alignment" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "Good evening. Tomorrow along the coast, e…",
"inputAudio": "https://your-cdn.com/input.mp3"
}
JSONOther models you might like
Explore similar audio & speech models.



