Forced Alignment Toolkit for Low-Resource Languages

This repository gathers tools and datasets to support building forced alignment systems for low-resource languages. Our current focus is producing grapheme-to-phoneme (G2P) models in IPA using Montreal Forced Aligner.

Languages

• Baca — collecting and curating lexicon data prior to MFA G2P training.
• Ewondo — preparing orthographic resources and cleaning pronunciation data.
• Fulfulde — scraping Webonary lexicons and converting entries to IPA.
• Wolof — adapting an existing MFA-compatible G2P model from the Kallaama dataset.

Getting Started

1. Install Montreal Forced Aligner following the official instructions.
2. Review each language folder for data preparation notes and scripts.
3. Use MFA to train or adapt language-specific IPA G2P models as resources become available.

Roadmap

• Gather and normalize lexicons for every target language.
• Train and validate MFA G2P models.
• Package pronunciation dictionaries for downstream forced alignment experiments.