@techreport{oai:ipsj.ixsq.nii.ac.jp:00190620,
 author = {太刀岡, 勇気 and Yuuki, Tachioka},
 issue = {7},
 month = {Jul},
 note = {音声区間検出を行う際には，パワーに基づく方法がよく使われる．しかしながらこの方法は高騒音下において性能の低下が著しいため，近年ではスペクトルの形状を考慮するような方法が提案されており，とりわけ深層神経回路網に基づく方法が性能が良いことが知られている．本報では，この方法の更なる改善を目的として，発話者の特徴や発話内容に応じた補助特徴量を用いる方法を提案する．特徴量として，非負値行列因子分解の活性化と音素ごとの事後確率を採用し，これらの有効性を車内環境での評価実験により確認した．, For voice activity detection, power-based methods are widely used; however, because these methods are susceptible to noise, recently, methods that consider the shape of spectrum have been proposed. In particular, deep neural network based methods have outperformed other methods. This paper aims to improve these methods by using auxiliary features that correspond to the speaker characteristics and the contents of the utterances. This paper proposes to use activation of non-negative matrix factorization and posterior probabilities of phonemes as an auxiliary feature and validates the effectiveness on the experiments in in-car environments.},
 title = {NMFと音響モデル併用型DNNに基づく音声区間検出},
 year = {2018}
}