A Hybrid Fundamental Frequency Estimator for Singing Voice
11:31 AM 8/7/99
Abstract
Fundamental frequency estimators are important in studying a singer's
performance. In this report, a hybrid fundamental frequency estimator is
developed to detect the variation of the fundamental frequency of singing
voices. The hybrid estimator combines the maximum likelihood harmonic
matching and the auto-correlation methods. Using the harmonic matching to
refine the search range for the auto-correlation method, we can reduce the
computation and achieve accurate estimates in fine resolutions. This
system has been tested on clean recordings of singing voices successfully.
Although only singing voices have been tested, the system should work on
any harmonic sounds.
For details, please see this
report.
Last modified: 11:32 AM 8/7/99
email: Hui-Ling Lu