An adaptive penalty approach to multi-pitch estimation

摘要: This work treats multi-pitch estimation, and in particular the common misclassification issue wherein pitch at half of true fundamental frequency, here referred to as a suboctave, is chosen instead pitch. Extending on current methods which use an extension Group LASSO for this introduces adaptive total variation penalty, both enforce group- block sparsity, deal with errors due sub-octaves. The method shown outperform state-of-the-art sparse methods, where model orders are unknown, while also requiring fewer tuning parameters than these. several conventional estimation even when these virtued oracle orders.