Naija Human vs Machine Transcription (Health Domain)

The Naija Machine vs Human Transcription in the Health Domain dataset contains 20 records documenting machine-generated and human-corrected transcriptions of audio content relating to health and healthcare. The dataset is provided in an OpenDocument Spreadsheet (.ods) file comprising three columns: ID, transcript, and corrected_transcript. The ID column identifies each record, transcript contains the initial or machine-generated transcription, and corrected_transcript contains the corresponding human-corrected version, based on the dataset description supplied by the creator. The transcripts cover health-related topics, including healthcare quality, fever, nursing, nutrition, cancer, and breastfeeding. The dataset may support research in automatic speech recognition, machine transcription evaluation, human post-editing, transcription quality assessment, and natural language processing for Nigerian Pidgin English in the health domain. Comparing the two transcription versions may help researchers examine differences in completeness, wording, and transcription accuracy. The accompanying README and data dictionary provide information about the file structure, variables, and relevant data-quality considerations. The original audio recordings are not included in the archive because permission to redistribute them is unavailable. Consequently, the dataset supports text-based analysis of the available transcripts but does not, by itself, permit direct acoustic evaluation of the machine-generated transcriptions against the source audio. Researchers should also note that the paired transcripts vary in length, and their correspondence should be examined carefully before calculating transcription error rates or conducting sentence-level comparisons.