blocker-abby/xh-en-health-data

Audio and bilingual text dataset for isiXhosa/English, presented in the CL4Health paper "Benchmarking IsiXhosa Automatic Speech Recognition and Machine Translation for Digital Health Provision" (Blocker et al., 2025) # isiXhosa/English Audiotextual Dataset - Health Domain This dataset contains bilingual audio and corresponding text (in isiXhosa and English). It is adapted from the PriMock57 dataset found here: github.com The repository contains three folders: - en audio: audio recordings in English - xh audio: audio recordings in isiXhosa - en text: audio transcriptions in English - xh text: audio transcriptions in isiXhosa Text data is in .txt format, separated by speaker. Audio data is in 48kHz monochannel FLAC format. Any queries can be sent to blcabb001@myuct.ac.za. Abby Blocker, Francois Meyer, Ahmed Biyabani, Joyce Mwangama, Mohammed Ishaaq Datay. 2025. "Benchmarking IsiXhosa Automatic Speech Recognition and Machine Translation for Digital Health Provision." CL4Health @ NAACL 2025.