Combining a recursive approach via non-negative matrix factorization and Gini index sparsity to improve reliable detection of wheezing sounds

•We propose a method to locate time intervals in which wheeze sounds are active.•A recursive orthogonal NMF and spectral sparsity provided by Gini index.•The spectral sparsity attempts to model the periodic nature shown by wheeze sounds.•The proposed method provides reliable and promising wheezing detection results.•The proposed method does not depend on any training dataset (unsupervised approach).