A Dataset for Multimodal Music Information Retrieval of Igbo Musical Videos

Music information retrieval has developed fast in recent times, but it is still mainly concerned with Western musical styles and languages, leading to no systematic answer for the valuable music from Africa. Out of all African countries, Igbo represents a unique musical culture but still receives less attention in global music. Therefore, this work uses a multimodal dataset made up of 85 authentic Igbo musical videos from YouTube, comprising 1,627 fifteen-second segments and 33,649 extracted video frames. The researchers gathered data in a detailed and organized way, including detailed cultural notes provided by Igbo speakers to make the study authentic. The recorded material covers twenty genres, ranging from highlife and traditional music to contemporary gospel and afrobeat, and also encompasses multiple Igbo dialects from different areas including Central Igbo, Onitsha, Owerri, and Ebonyi. By applying audio processing methods, 39 features were collected, for instance MFCCs, chroma vectors and spectral characteristics, together with in-depth metadata covering emotional analysis, translations of the lyrics and explanations of cultural influences. This work addresses a major lack of data on African languages in music research and acts as a digital library for Igbo music, so experts can design smart systems and support technological progress in African music. Keywords: Multimodal, Music information retrieval, Igbo dialects, African music, Acoustic features, Sentiment analysis, Cultural preservation