Kaosarat-Aina/Yoruba-learner-speech-corpus

# Project Title # Yoruba Learner Speech Corpus (YLSC) _The first phonemically and tonally annotated speech corpus of Yoruba learner pronunciation designed for linguistic research and CAPT development._ The Yoruba learner Speech Corpus is a pioneering, phonemically and tonally annotated dataset designed to support research in Second Language Acquisition (SLA), Automatic Speech Recognition (ASR), and Computer-Assisted Pronunciation Training (CAPT). By capturing authentic learner speech with expert phonetic diagnostics, this project addresses the significant gap in resources for tonal African languages. **Project lead: Kaosarat Aina**, **Indiana University Bloomington** **Contact: kbaina@iu.edu** # Overview The Yoruba Learner Speech Corpus (YLSC) is a developing dataset of speech produced by second-language and heritage learners of Yoruba. Learning a second language (L2) is a complex process where speaking proficiency often serves as the foundation for learner confidence. However, learners frequently lack immediate corrective feedback during self-study. While commercial CAPT tools exist, their proprietary nature limits research and modification. This corpus provides an open-source foundational resource for developing robust, diagnostic feedback systems. # Research Motivation * **Tonal Complexity**: Yoruba's three-way tonal contrast (High, Mid, and Low) presents a steep challenge for L2 learners, particularly those from non-tonal L1 backgrounds. * **Segmental Challenges**: Beyond tones, learners often struggle with unique Yoruba phonemes, such as co-articulated labial-velar stops ([ɡb], [kp]) and the Advanced Tongue Root (±ATR) vowel harmony system. * **Native vs. Learner Data**: Existing Yoruba corpora are designed for native speech, lacking the specific phonological errors—such as vowel harmony violations or tonal misassignments—inherent to learners. * **Diagnostic Gaps**: Current research requires error-annotated corpora to move beyond simple "pass/fail" error …