Artificial Intelligence for Cross-Cultural Percussion Analysis: A PRISMA-Based Systematic Review of Tabla and Conga

While AI has enhanced automatic drum transcription, culturally unique percussion instruments, due to their common peculiar timbral, harmonic and microtiming properties, are still underrepresented. This systematic review explores the analysis of Hindustani tabla and Afro-Cuban conga (and other related non-Western percussion traditions) using an AI approach. The peer-reviewed papers retrieved from IEEE Xplore, ACM Digital Library, Scopus, ISMIR and Google Scholar according to PRISMA 2020 guidelines. Fourty-seven studies were synthesized across the areas of transcription, rhythmic analysis, low resource learning, and generative modelling of adequate quality and the author applied certain preceding eligibility criteria. Convolutional recurrent neural networks (CRNNs) were able to distinguish between tableau and tabla-strokes in concert recordings with an F score of 0.965, while the other deep-learning models got an accuracy of 97.50% in rhythmic-pattern recognition. The groove modelling of Afro-Latin showed that the adaptability under limited annotated data was boosted by meta-learning and data augmentation, while tatum-level beat and microtiming analysis was critical. The generation of stylistic percussion patterns is another major challenge in the generative approach, which is also faced by sparse datasets, domain shift, cultural bias and real time polyphonic transcription. Therefore, for accurate and authentic cross-cultural percussion analysis culturally informed, transparent and human-centred models are crucial for AI.