US 12,406,657 B2
Adaptable acoustic model built with limited labeling data
Zhong Fang Yuan, Xi'an (CN); Si Tong Zhao, Beijing (CN); Tong Liu, Xi'an (CN); Yi Chen Zhong, Shanghai (CN); and Yuan Yuan Ding, Shanghai (CN)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed on Feb. 10, 2023, as Appl. No. 18/167,127.
Prior Publication US 2024/0274125 A1, Aug. 15, 2024
Int. Cl. G10L 15/08 (2006.01); G10L 15/06 (2013.01)
CPC G10L 15/063 (2013.01) [G10L 2015/0638 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A processor-implemented method for building an acoustic model, the method comprising:
performing contrastive pre-training of the acoustic model;
building a dataset classifier using prompt engineering, wherein the prompt engineering comprises using one or more prompt templates to convert each of one or more received labels into one or more text descriptions;
performing a prediction process; and
performing zero-shot audio prediction using the pre-trained acoustic model and the one or more text descriptions.