These are Nigerian music benchmark datasets curated for Music Information Retrieval (MIR) tasks and the research is funded by the Black in AI Research (BlackAIR) grant for summer from Jul-Aug, 2021

These datasets were curated gradually by myself (Dr Sakinat Folorunso), my undergraduate student (Owodeyi Adeoye Bashir), and my M. Sc. student (Odumosu Adesola) for research purposes only (and obviously no copyright permission). The curation of the audio files was started in 2019-2021 from a variety of sources including MPEG Audio Layer-3/4(MP3/MP4) with the best quality from YouTube using vidmate and 4k video downloader. However, some of the songs file were collected from Disc Jockeys (DJs), personal CDs, radio, microphone recordings, in order to represent a variety of recording conditions.

English Contemporary Songs Dataset

The dataset consists of 712 audio tracks each 30 seconds long. It contains 8 genres, each represented by 100 or fewer tracks. The tracks are all 22050Hz Mono 16-bit audio files in .wav format.

The link to download the dataset will appear here shortly

Traditional Songs Dataset

The dataset consists of 478 audio tracks each 30 seconds long. It contains 5 genres, each represented by 100 or fewer tracks. The tracks are all 22050Hz Mono 16-bit audio files in .wav format.

NaijArtist Dataset

There is a scarcity of Nigerian song Artists for MIR tasks and research. Hence, it is useful and important to create a benchmark dataset and task that researchers could download and run on their own machines. The sole aim of the NaijArtist music dataset is for research purposes only.

NaijArtist is a database consisting of 30 seconds of 30 differents tracks by each of 130 artists, making a total of 1,413 tracks. This dataset was born out of our work in music genre classification (Folorunso et al, 2021) and research further into artist recognition and identification, where we identified 130 Artists.

These audio files can be freely downloaded at Mendeley in .wav format at 22kHz.