はじめにKaggle Free Sound Audio Tagging 2019で学ぶ音声処理ではKaggleコンペとその解法を題材に音声処理について解説しています。この記事は、メルスペクトログラムの計算中に出てきたメルフィルタバンクについて解説します。 librosa.feature.melspectrogramlibrosa.feature.melspectrogramのコードを読むで出てきた
Each filter in the bank is a bandpass raised cosine filter centered at frequency ω i (corresponding to a resolution ω i = 1 / R i ), see Supplementary Material, and represented by matrix H i . In this way, the local filter is constructed by the addition of many bandpass filters with different weights covering the frequency range up to ...