Publications
OTHERS (DOMESTIC) Raw or cooked? That is the Question in Adaptive Noise Cancelling
Akihiko Sugiyama
電子情報通信学会第37回信号処理シンポジウム (SIPシンポジウム)
December 13, 2022
This paper presents how to use obtained data, raw or cooked, in a scenario of adaptive noise cancellation. Two types of data are generated as estimated SNRs which are used to control the adaptation stepsize of the adaptive filter therein. They are different in the method of generation and in the statistical characteristics like ingredients of a dish. The way of cooking differs from one ingredient to another. Similarly, in adaptive noise cancellation, the two types of data are investigated in the way they are processed. It is shown, like fish fillet, that good data with stable quality can be served raw whereas data with considerable variance in quality is better to be served cooked, which is data-dependent cooking. Evaluations with clean speech and noise recorded at a busy station demonstrate that data-dependent cooking provides as much as 6 dB better result.