Tonguscape Dataset Builder

This is a dataset builder for Tonguscape introduced in Tonguescape: Exploring Language Models Understanding of Vowel Articulation (NAACL 2025).

This dataset is segmented by phoneme from rtMRIDB. The phoneme is defined in original DB.

Directory Structure

• rmri-extract/: The directory for extracting the base of this dataset from rtMRIDB
• scripts/: The directory for the dataset builder scripts (e.g., removing audio or overlaying shapes)
• data/: The directory for the dataset index
• assets/: The directory for the dataset annotations