US 11,862,183 B2
Methods of encoding and decoding audio signal using neural network model, and devices for performing the methods
Jongmo Sung, Daejeon (KR); Seung Kwon Beack, Daejeon (KR); Mi Suk Lee, Daejeon (KR); Tae Jin Lee, Daejeon (KR); Woo-taek Lim, Daejeon (KR); and Inseon Jang, Daejeon (KR)
Assigned to Electronics and Telecommunications Research Institute, Daejeon (KR)
Filed by Electronics and Telecommunications Research Institute, Daejeon (KR)
Filed on Jul. 6, 2021, as Appl. No. 17/368,390.
Claims priority of application No. 10-2020-0083075 (KR), filed on Jul. 6, 2020; and application No. 10-2021-0049753 (KR), filed on Apr. 16, 2021.
Prior Publication US 2022/0005487 A1, Jan. 6, 2022
Int. Cl. G10L 19/032 (2013.01)
CPC G10L 19/032 (2013.01) 6 Claims
OG exemplary drawing
 
1. An encoding method for audio signal comprising:
computing first feature information of an input audio signal inputting the input audio signal to a recurrent encoding model;
generating a first bitstream by quantizing the first feature information;
generating a first output signal from a quantized first feature information by inputting the quantized first feature information a recurrent decoding model;
calculating a residual signal by subtracting the first output signal from the input audio signal;
computing second feature information of the residual signal by inputting the residual signal to a nonrecurrent encoding model;
generating a second bitstream by quantizing the second feature information; and
outputting an overall bitstream by multiplexing the first bitstream and the second bitstream.