Penjernihan Derau pada Suara Kanal Tunggal dengan Pembelajaran Faktorisasi Matriks Non-negatif tanpa Pengawasan

This article examines an approach of denoising method on single channel using Non-negative Matrix Factorization (NMF) on unsupervised-learning scheme. This technique utilizes the property of NMF which unravels spectrogram matrices of noise-interfered speech and noise itself into their building-block vector. As extension for NMF, Wiener filter is applied in the end of steps. This method is designated to run in low latency system, hence preparing certain noise model for particular condition beforehand is impractical. Thus the noise model is taken automatically from the unvoiced part of noise-interfered speech. The contribution achieved in this research is the kind of NMF learning using linear and non-linear constraint which is done without explicitly providing noise models. Therefore the denoising process could be undergone flexibly in any noise condition.

Abstract licence: CC BY-NC-SA

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.