Naija Audio Data

The Naija Audio Transcription Dataset comprises five WAV audio items accompanied by transcription and annotation data in JSON format. The JSON file, Trancribed Naija Audio Data.json, contains the full transcription associated with each audio item, language and acoustic labels, annotator notes where available, and annotation metadata. The transcription text is written in Latin script as supplied in the dataset. All five records contain a full-transcription field, a language selection recorded as Other (note in comment), and an acoustic label of Mobile — minor noise. Four records include notes stating that the language is Naija and that the original data provided was translated into Naija because data in Naija was not available. The JSON documents this translation-related step but does not fully specify the original source material or the process used to create the audio. Two of the five records contain time-aligned annotations, including segment start and end times, channel information, speech labels, and associated text linked to particular intervals. However, time-aligned annotations are not available for every audio item, and segment and text entries are not consistently paired throughout the JSON. Users requiring complete word-level or time-aligned transcriptions should review the annotations before using the dataset for such purposes. The JSON also contains task identifiers, annotation identifiers, timestamps, and other annotation-tool metadata. One record includes a draft transcription that represents the year in words, while the saved transcription uses numerals. The saved transcription is the primary recorded annotation; the status of the draft should be confirmed against the project’s annotation workflow where relevant. The dataset may be useful for linguistic research involving Naija transcription and for developing or testing speech and language processing workflows. Its suitability for specific applications should be assessed in light of the partial time-aligned annotations and the provenance and methodological information not documented in the JSON.