Designing A Korean French-Learners’ Speech Corpus (KFLSC) for Spoken Language Assessment

This paper presents how to design a French speech corpus produced by Korean learners (Korean French-learners’ Speech Corpus: KFLSC), which aims to provide an AI training database for spoken language assessment within the Korean AI Hub Project. The corpus is required to include audio recording, their corresponding transcription, metadata, assessment labeling, and error annotations. Both reading speech and spontaneous speech will be recorded by using our scripts and instruction prompts. The transcription of audio files will be performed both automatically and manually. Detailed speech assessment and scoring information are provided along with proposed pronunciation and spoken language assessment rubrics. We also suggest tagging guidance on pronunciation and spoken language errors. This study will contribute to building a large-scale French learner corpus for French spoken language assessment and the education of French as a foreign language, and to the development of automatic speech assessment and computer-aided language learning technologies.