METHOD FOR ENHANCING TELEPHONE VOICE SIGNAL BASED ON CONVOLUTIONAL NEURAL NETWORK

METHOD FOR ENHANCING TELEPHONE VOICE SIGNAL BASED ON CONVOLUTIONAL NEURAL NETWORK

To provide a method for reducing effects of acoustic distortion on telephone voice based on a convolutional neural network (CNN).SOLUTION: A method includes: a preprocessing step A which involves extracting an amplitude and a phase of spectral representation of a telephone voice signal; a voice enha...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: INIIGO GARCIA MORTE, ANTONIO MIGUEL ARTIAGA, JAVIER GALLART MAURI, ALFONSO ORTEGA GIMENEZ, EDUARDO LLEIDA SOLANO, DAYANA RIBAS GONZALEZ
Format: Patent
Sprache:eng ; jpn
Schlagworte:
ACOUSTICS
MUSICAL INSTRUMENTS
PHYSICS
SPEECH ANALYSIS OR SYNTHESIS
SPEECH OR AUDIO CODING OR DECODING
SPEECH OR VOICE PROCESSING
SPEECH RECOGNITION
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:To provide a method for reducing effects of acoustic distortion on telephone voice based on a convolutional neural network (CNN).SOLUTION: A method includes: a preprocessing step A which involves extracting an amplitude and a phase of spectral representation of a telephone voice signal; a voice enhancement step B which estimates a Wiener gain by a trained convolutional neural network, and emphasizes the amplitude of the spectral representation of the voice signal using a voice enhancement filter based on the estimated Wiener gain; and a post-processing step C which obtains the enhanced voice signal based on the amplitude and initial phase of the emphasized spectral representation.SELECTED DRAWING: Figure 2 【課題】畳み込みニューラルネットワーク（ＣＮＮ）に基づいて電話音声における音響歪みの影響を低減する方法を提供する。【解決手段】方法は、電話音声信号のスペクトル表現の振幅及び位相を抽出することを含む前処理段階Ａと、訓練された畳み込みニューラルネットワークによりウィーナーゲインを推定し、推定されたウィーナーゲインに基づく音声強調フィルタにより音声信号のスペクトル表現の振幅を強調する音声強調段階Ｂと、強調されたスペクトル表現の振幅と初期位相とに基づき強調された音声信号を得る後処理段階Ｃと、を含む。【選択図】図２