US 12,334,080 B2
Neural network-based signal processing apparatus, neural network-based signal processing method, and computer-readable storage medium
Qiongqiong Wang, Tokyo (JP); Takafumi Koshinaka, Tokyo (JP); and Kong Aik Lee, Tokyo (JP)
Assigned to NEC CORPORATION, Tokyo (JP)
Appl. No. 17/764,291
Filed by NEC Corporation, Tokyo (JP)
PCT Filed Oct. 18, 2019, PCT No. PCT/JP2019/041226
§ 371(c)(1), (2) Date Mar. 28, 2022,
PCT Pub. No. WO2021/075063, PCT Pub. Date Apr. 22, 2021.
Prior Publication US 2022/0335950 A1, Oct. 20, 2022
Int. Cl. G10L 17/18 (2013.01); G06N 3/08 (2023.01); G10L 17/02 (2013.01)
CPC G10L 17/18 (2013.01) [G06N 3/08 (2013.01); G10L 17/02 (2013.01)] 12 Claims
OG exemplary drawing
 
1. A neural network-based signal processing apparatus comprising:
at least one memory storing instructions; and
at least one processor configured to execute the instructions to:
receive multi-dimension features containing two or more two-dimension feature maps, where the multi-dimension features are labeled;
produce a first weight matrix including a plurality of first attentive weights for a plurality of elements, respectively, in the multi-dimension features by using a time-attentive neural network;
produce a second weight matrix including a plurality of second attentive weights for the plurality of elements, respectively, in the multi-dimensional features by using a channel-frequency-attentive neural network;
produce low-dimension features or posterior probabilities for designated classes, based on the multi-dimension features and the first and second attention weights;
multiply the first and second weight matrices and the multi-dimensional features;
train an attention network jointly with a classification network, using the multi-dimension features as labeled and after multiplication;
receive input data; and
classify the input data using the attention network and the classification network that have been trained.