TY - GEN
T1 - Noisy-target Training
T2 - 29th European Signal Processing Conference, EUSIPCO 2021
AU - Fujimura, Takuya
AU - Koizumi, Yuma
AU - Yatabe, Kohei
AU - Miyazaki, Ryoichi
N1 - Publisher Copyright:
© 2021 European Signal Processing Conference. All rights reserved.
PY - 2021
Y1 - 2021
N2 - Deep neural network (DNN)-based speech enhancement ordinarily requires clean speech signals as the training target. However, collecting clean signals is very costly because they must be recorded in a studio. This requirement currently restricts the amount of training data for speech enhancement to less than 1/1000 of that of speech recognition which does not need clean signals. Increasing the amount of training data is important for improving the performance, and hence the requirement of clean signals should be relaxed. In this paper, we propose a training strategy that does not require clean signals. The proposed method only utilizes noisy signals for training, which enables us to use a variety of speech signals in the wild. Our experimental results showed that the proposed method can achieve the performance similar to that of a DNN trained with clean signals.
AB - Deep neural network (DNN)-based speech enhancement ordinarily requires clean speech signals as the training target. However, collecting clean signals is very costly because they must be recorded in a studio. This requirement currently restricts the amount of training data for speech enhancement to less than 1/1000 of that of speech recognition which does not need clean signals. Increasing the amount of training data is important for improving the performance, and hence the requirement of clean signals should be relaxed. In this paper, we propose a training strategy that does not require clean signals. The proposed method only utilizes noisy signals for training, which enables us to use a variety of speech signals in the wild. Our experimental results showed that the proposed method can achieve the performance similar to that of a DNN trained with clean signals.
KW - Deep neural network (DNN)
KW - Noise2Noise
KW - Single-channel speech enhancement
KW - Training target
UR - http://www.scopus.com/inward/record.url?scp=85114581068&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85114581068&partnerID=8YFLogxK
U2 - 10.23919/EUSIPCO54536.2021.9616166
DO - 10.23919/EUSIPCO54536.2021.9616166
M3 - Conference contribution
AN - SCOPUS:85114581068
T3 - European Signal Processing Conference
SP - 436
EP - 440
BT - 29th European Signal Processing Conference, EUSIPCO 2021 - Proceedings
PB - European Signal Processing Conference, EUSIPCO
Y2 - 23 August 2021 through 27 August 2021
ER -