US 12,148,441 B2
Source separation for automatic speech recognition (ASR)
Alon Slapak, Mazor (IL); and Dani Cherkassky, Mazor (IL)
Assigned to KARDOME TECHNOLOGY LTD., Mazor (IL)
Appl. No. 17/437,748
Filed by Kardome Technology Ltd., Mazor (IL)
PCT Filed Mar. 10, 2019, PCT No. PCT/IB2019/051933
§ 371(c)(1), (2) Date Sep. 9, 2021,
PCT Pub. No. WO2020/183219, PCT Pub. Date Sep. 17, 2020.
Prior Publication US 2022/0148611 A1, May 12, 2022
Int. Cl. G10L 21/0232 (2013.01); G10L 15/22 (2006.01); G10L 21/0208 (2013.01); G10L 21/0216 (2013.01); G10L 21/0264 (2013.01)
CPC G10L 21/0232 (2013.01) [G10L 15/22 (2013.01); G10L 21/0264 (2013.01); G10L 2021/02087 (2013.01); G10L 2021/02166 (2013.01)] 23 Claims
OG exemplary drawing
 
1. An apparatus comprising:
an input to receive a plurality of input signals representing an acoustic input;
an output to provide a plurality of output signals configured for an Automatic Speech Recognition (ASR) process at a speech recognition engine; and
a source separator configured to generate the plurality of output signals based on the plurality of input signals and an error input, the error input based on feedback from the speech recognition engine, the error input representing an error of the speech recognition engine based on a plurality of previous output signals generated by the source separator and provided from the output to the speech recognition engine, wherein the source separator is configured to generate the plurality of output signals by processing the plurality of input signals to separate the acoustic input into a plurality of source-separated acoustic signals corresponding to a plurality of acoustic sources, respectively.