Fact · Transcription
The Montreal Forced Aligner was evaluated on word and phone boundaries against boundaries that humans had marked by hand.
Checked against its source on October 17, 2026
In the source's own words
“relative to human-annotated boundaries”
Montreal Forced Aligner: Trainable Text-Speech Alignment Using Kaldi, Michael McAuliffe et al., Proceedings of Interspeech 2017, 2017, Abstract.
Open the original (DOI 10.21437/Interspeech.2017-1386)- What it measures
- Forced alignment matches a known transcript to audio. Its quality is a boundary error, which is why the field measures it against human marks.
- What it does not show
- The abstract describes the evaluation method. The numeric error figures are in the full paper.