Forced Alignment Voiceover Analysis

In courseware development, we often find last-minute errors in the AI speech-to-text. We review. We listen. We review more. But errors slip through. Much of our courseware is over highly technical topics and full of terms and acronyms, so tedious to review.

One avenue we are investigating is the use of Forced Alignment tools to provide an analysis of the voiceovers and call out any errors found. Then break the spoken word into phonemes, based on the written script. Then provide a comparison of the script vs the transcript.

Has anyone been using these tools? Looking for feedback on ease of use or any training.

Looking forward to comments or suggestions!

Travis

Tool Name	Description	Download Link
Montreal Forced Aligner (MFA)	Trainable text-speech alignment using Kaldi; supports phoneme-level alignment and multilingual models.	https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner
Gentle	Lightweight forced aligner for English; easy to use for word-level alignment.	https://github.com/lowerquality/gentle
Penn Phonetics Lab Forced Aligner (P2FA)	Classic aligner for English; requires HTK installation.	https://www.ling.upenn.edu/phonetics/p2fa/
Prosodylab-Aligner	Designed for laboratory speech; integrates with Praat for prosody analysis.	http://www.prosodylab.org/lab/software.html
FAVE-align	Sociolinguistic alignment tool; based on P2FA.	https://fave.ling.upenn.edu/
Aeneas	Aligns audio and text for eBooks and captions; Python-based.	https://github.com/readbeyond/aeneas
Pyfoal	Python suite integrating MFA, P2FA, and RAD-TTS for alignment and ML tasks.	https://github.com/maxrmorrison/pyfoal
mms-300m-1130-forced-aligner	Multilingual forced aligner based on Meta’s wav2vec2; supports 158 languages.	https://sourceforge.net/projects/mms-300m-1130-forced-aligner/
Online Forced Aligner (UWM)	Web-based interface for Penn Phonetics Lab aligner; outputs Praat TextGrid.	https://web.uwm.edu/forced-aligner/