ANPLers at IqraEval Shared task: Adapting Whisper-large-v3 as Speech-to-Phoneme for Qur’anic Recitation Mispronunciation Detection
Year: 2025
Type: conference-paper
Abstract: Mispronunciation detection at the phoneme level provides detailed feedback for Quranic reciters.Standard speech-to-text models cannot capture subtle differences in letter pronunciation; thus, developing a speech-to-phoneme system is essential.Prior works have mainly explored encoder-only models.In this work, we adapt Whisper-large-v3 on the IqraEval dataset.Our experimental results show that the proposed system achieved an F1score of 0.3224, an accuracy of 0.6894, and a high recall of 0.7624.These results highlight promising directions for further research and development in phoneme-level mispronunciation detection.
Institutions: Prince Sultan University, Tuwaiq Academy
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037771