A neural network-supported two-stage algorithm for lightweight dereverberation on hearing devices

EURASIP Journal on Audio, Speech, and Music Processing

Empirical Research

Raphael Koning¹  Joachim Thiemann¹  Jean-Marie Lemercier²  Timo Gerkmann² 

[1] Advanced Bionics, Hannover, Germany;Signal Processing, Universität Hamburg, Hamburg, Germany;

关键词: Dereverberation;    Neural network;    End-to-end learning;    Hearing devices;   

DOI  :  10.1186/s13636-023-00285-8

 received in 2022-11-24, accepted in 2023-04-11,  发布年份 2023

来源: Springer

【 摘 要 】

A two-stage lightweight online dereverberation algorithm for hearing devices is presented in this paper. The approach combines a multi-channel multi-frame linear filter with a single-channel single-frame post-filter. Both components rely on power spectral density (PSD) estimates provided by deep neural networks (DNNs). By deriving new metrics analyzing the dereverberation performance in various time ranges, we confirm that directly optimizing for a criterion at the output of the multi-channel linear filtering stage results in a more efficient dereverberation as compared to placing the criterion at the output of the DNN to optimize the PSD estimation. More concretely, we show that training this stage end-to-end helps further remove the reverberation in the range accessible to the filter, thus increasing the early-to-moderate reverberation ratio. We argue and demonstrate that it can then be well combined with a post-filtering stage to efficiently suppress the residual late reverberation, thereby increasing the early-to-final reverberation ratio. This proposed two-stage procedure is shown to be both very effective in terms of dereverberation performance and computational demands, as compared to, e.g., recent state-of-the-art DNN approaches. Furthermore, the proposed two-stage system can be adapted to the needs of different types of hearing-device users by controlling the amount of reduction of early reflections.

【 授权许可】

CC BY   
© The Author(s) 2023

【 图 表 】

Fig. 4

40517_2023_256_Article_IEq150.gif

Fig. 28

Fig. 1

Fig. 27

Fig. 3

Fig. 26

Fig. 2

Fig. 25

Fig. 24

Fig. 22

Fig. 1

Fig. 3

40517_2023_256_Article_IEq140.gif

Fig. 3

40517_2023_256_Article_IEq139.gif

Fig. 1

Fig. 3

Fig. 1

40517_2023_256_Article_IEq135.gif

41512_2023_147_Article_IEq110.gif

Fig. 4

40517_2023_256_Article_IEq121.gif

40517_2023_256_Article_IEq120.gif

Fig. 18

40517_2023_256_Article_IEq119.gif

40517_2023_256_Article_IEq118.gif

Fig. 17

40517_2023_256_Article_IEq116.gif

40517_2023_256_Article_IEq115.gif

40517_2023_256_Article_IEq114.gif

40517_2023_256_Article_IEq113.gif

Fig. 14

Fig. 1

40517_2023_256_Article_IEq110.gif

40517_2023_256_Article_IEq109.gif

40517_2023_256_Article_IEq108.gif

40517_2023_256_Article_IEq107.gif

Fig. 4

40517_2023_256_Article_IEq106.gif

Fig. 2

Fig. 3

Fig. 2

40517_2023_256_Article_IEq99.gif

Fig. 3

Fig. 7

Fig. 1

Fig. 11

Fig. 10

Fig. 1

40517_2023_256_Article_IEq74.gif

Fig. 9

40517_2023_256_Article_IEq73.gif

Fig. 9

Fig. 8

Fig. 5

Fig. 3

Fig. 7

Fig. 1

Fig. 6

Fig. 2

Fig. 1

Fig. 1

Fig. 5

Fig. 4

Fig. 3

Fig. 2

Fig. 2

40517_2023_256_Article_IEq51.gif

Fig. 2

40517_2023_258_Article_IEq137.gif

40517_2023_258_Article_IEq135.gif

40517_2023_258_Article_IEq133.gif

Fig. 1

Fig. 1

40517_2023_256_Article_IEq14.gif

Fig. 6

Fig. 7

40517_2023_256_Article_IEq11.gif

Fig. 5

40517_2023_256_Article_IEq5.gif

40517_2023_256_Article_IEq4.gif

41116_2023_36_Article_IEq117.gif

Fig. 2

41116_2023_36_Article_IEq116.gif

41116_2023_36_Article_IEq818.gif

41116_2023_36_Article_IEq816.gif

41116_2023_36_Article_IEq813.gif

41116_2023_36_Article_IEq811.gif

40517_2023_258_Article_IEq102.gif

41116_2023_36_Article_IEq806.gif

41116_2023_36_Article_IEq803.gif

41116_2023_36_Article_IEq801.gif

41116_2023_36_Article_IEq799.gif

41116_2023_36_Article_IEq797.gif

Fig. 1

41116_2023_36_Article_IEq791.gif

41116_2023_36_Article_IEq786.gif

41116_2023_36_Article_IEq785.gif

41116_2023_36_Article_IEq781.gif

41116_2023_36_Article_IEq757.gif

41116_2023_36_Article_IEq689.gif

41116_2023_36_Article_IEq688.gif

41116_2023_36_Article_IEq686.gif

41116_2023_36_Article_IEq684.gif

41116_2023_36_Article_IEq682.gif

41116_2023_36_Article_IEq680.gif

Fig. 1

Fig. 1

40517_2023_258_Article_IEq25.gif

41116_2023_36_Article_IEq675.gif

41116_2023_36_Article_IEq674.gif

41116_2023_36_Article_IEq672.gif

41116_2023_36_Article_IEq670.gif

41116_2023_36_Article_IEq667.gif

41116_2023_36_Article_IEq664.gif

41116_2023_36_Article_IEq663.gif

41116_2023_36_Article_IEq658.gif

41116_2023_36_Article_IEq657.gif

41116_2023_36_Article_IEq656.gif

41116_2023_36_Article_IEq655.gif

41116_2023_36_Article_IEq653.gif

41116_2023_36_Article_IEq649.gif

41116_2023_36_Article_IEq644.gif

Fig. 2

41116_2023_36_Article_IEq503.gif

41116_2023_36_Article_IEq502.gif

41116_2023_36_Article_IEq501.gif

Fig. 7

Fig. 2

Fig. 1

Fig. 2

40517_2023_258_Article_IEq8.gif

40517_2023_258_Article_IEq7.gif

Fig. 1

41116_2023_36_Article_IEq499.gif

41116_2023_36_Article_IEq497.gif

41116_2023_36_Article_IEq496.gif

Fig. 1

41116_2023_36_Article_IEq493.gif

12888_2023_4776_Article_IEq1.gif

41116_2023_36_Article_IEq489.gif

Fig. 1

41116_2023_36_Article_IEq372.gif

Fig. 1

41116_2023_36_Article_IEq303.gif

41116_2023_36_Article_IEq280.gif

41116_2023_36_Article_IEq267.gif