| Summary: | 碩士 === 國立交通大學 === 電信研究所 === 83 === HRTFs are the transfer functions from 3-D positions to both
ears. A total of 710 HRTFs, measured from the dummy head at the
MIT Media Lab, constituted the data set to be processed. This
thesis describes clustering and PCA approaches to simplify HRTF
clustering aims to choose some most significant HRTFs among the
whole data set. We propose to use cepstrum clustering, as
opposed to uniform to achieve lower mismatch error. The essence
of the PCA algorithm is to search for some basic so that the
attributes of HRTFs are the combination of these. We will
discuss and compare three PCA algorithms (LM-PCA, M-PCA, and
the QR method with pivoting. Another merit of the PCA is to
interpolate new HRTFs which are excluded in the data set.
Listening tests of a moving sound source are also made to
justify these algorithms.
|
|---|
