An improvement of the eCATE algorithm for F0 detection
Résumé
In this paper, we propose a modified version of the eCATE algorithm for F0 detection, which we affirm to be robust against changing databases. The improvement in the V/UV decision we made is relative to the use of different weighting functions applied to the CATE signal giving different possible F0 values and to a majority vote algorithm which selects the elected F0 index candidate. Although a lot of pitch detection algorithms have been proposed in the literature, F0 detection remains an open research problem. The results of our method using the Bagshaw database of the Center for Speech Technology Research at Edinburgh, exhibit important improvement compared to other published detection algorithms using this same database.