NOISE SUPPRESSION DEVICE, NOISE SUPPRESSION METHOD AND AUDIO INPUT EQUIPMENT

To provide a noise suppression device capable of eliminating a time lag when using conventional time frequency masking.SOLUTION: A noise suppression device (1) comprises: a Fourier transformation unit (12) which transforms an acquired input signal into a first power spectrum; an auto-correlation est...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: HARUTA CHIHO, ONO JUNKI, HARUHARA MASAHIRO
Format: Patent
Sprache:eng ; jpn
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:To provide a noise suppression device capable of eliminating a time lag when using conventional time frequency masking.SOLUTION: A noise suppression device (1) comprises: a Fourier transformation unit (12) which transforms an acquired input signal into a first power spectrum; an auto-correlation estimation unit (17) which uses a neural network to estimate a first auto-correlation corresponding to an unacquired frame of the input signal and a second auto-correlation corresponding to a desired signal having a noise component of the unacquired frame suppressed; a filter coefficient update unit (15) which updates a filter coefficient for suppressing the noise component based upon the first/second auto-correlations; a convolutional operation unit (16) which uses the previously updated filter coefficient to perform a convolutional operation on the acquired input signal when the unacquired frame is input; a window function multiplication unit (18); and an overlap addition unit (19) which generates the desired signal.SELECTED DRAWING: Figure 6 【課題】従来の時間周波数マスキングを用いる場合の時間遅延を解消することが可能な雑音抑制装置を提供する。【解決手段】雑音抑制装置(1)は、取得済入力信号を第1パワースペクトルに変換するフーリエ変換部(12)と、ニューラルネットワークを用いて入力信号の未取得フレームに対応する第1自己相関及び未取得フレームの雑音成分が抑制された所望信号に対応する第2自己相関を推定する自己相関推定部(17)と、第1/第2自己相関に基づいて雑音成分を抑制するためのフィルタ係数を更新するフィルタ係数更新部(15)と、未取得フレームが入力されると予め更新されたフィルタ係数を用いて取得済入力信号に畳み込み演算を行う畳み込み演算部(16)と、窓関数乗算部(18)と、所望信号を生成するオーバーラップ加算部(19)とを備えている。【選択図】図6