Comparison of Four Polynomial Kernels for Enhancement of Autocorrelation-Based Pitch Estimates

Hee-Suk PANG  Byeong-Moon JEON  

Publication
IEICE TRANSACTIONS on Fundamentals of Electronics, Communications and Computer Sciences   Vol.E87-A   No.9   pp.2459-2462
Publication Date: 2004/09/01
Online ISSN: 
DOI: 
Print ISSN: 0916-8508
Type of Manuscript: LETTER
Category: Engineering Acoustics
Keyword: 
autocorrelation,  pitch estimation,  polynomial interpolation,  

Full Text: PDF(165.3KB)>>
Buy this Article




Summary: 
Whereas the autocorrelation is frequently used for pitch estimation, the resultant estimates usually suffer from inaccuracy. Instead of upsampling, we can improve the accuracy of the estimates by applying polynomial interpolation to the autocorrelation directly. For that purpose, four kernels, which are interpolating quadratic, quadratic-B spline, cubic-B spline, and cubic convolution kernels respectively, have been compared. Experiments show that the cubic B spline kernel shows the best performance, a little inferior to the computationally intensive upsampling procedure. The quadratic B spline kernel shows also reasonable performance with the merit of the further reduced computational complexities compared with the cubic B spline kernel.