NONLINEAR SPECTRAL SUBTRACTION BERBASIS TSALLIS STATISTICS UNTUK PENINGKATAN KUALITAS SINYAL UCAPAN
Abstract: The presence of the
noise degrades the quality and the intelligibility of the speech signal and
hence reduces the performance of speech based application. Spectral subtraction
is a popular method to remove additive noise. However, it has a major shortcoming
of introducing musical noise. Several variants of spectral subtraction have
been proposed to tackle this issue. One of them is the introduction of
oversubtraction factor in the spectral subtraction formula. This approach
nonlinear spectral subtraction. However, this factor is decided heurestically.
Tsallis statistics has found to introduce a nonlinear subtraction naturally. A
new variant of spectral subtraction, which is called q-spectral subtraction,
has been derived. q-SS has been found to be effective for improving the
robustness of speech recognition performance against noise. However, the
evaluation of this method for speech enhancement tasks has not been explored
yet. In this paper, the performance of q-spectral subtraction for speech enhancement
task is investigated. It is found that q-SS is better than other spectral
subtraction methods in improving the quality of speech signals.
Penulis: Hilman Ferdinandus
Pardede
Kode Jurnal: jptkomputerdd150465
