US 11,776,535 B2
Semantic understanding method and apparatus, and device and storage medium
He Zhang, Beijing (CN); Hang Li, Beijing (CN); and Yang Wang, Beijing (CN)
Assigned to BEIJING BYTEDANCE NETWORK TECHNOLOGY CO., LTD., Beijing (CN)
Filed by BEIJING BYTEDANCE NETWORK TECHNOLOGY CO., LTD., Beijing (CN)
Filed on Aug. 11, 2022, as Appl. No. 17/885,965.
Application 17/885,965 is a continuation of application No. PCT/CN2021/084846, filed on Apr. 1, 2021.
Claims priority of application No. 202010356038.8 (CN), filed on Apr. 29, 2020.
Prior Publication US 2022/0392440 A1, Dec. 8, 2022
Int. Cl. G10L 15/18 (2013.01); G10L 15/183 (2013.01)
CPC G10L 15/1815 (2013.01) [G10L 15/183 (2013.01)] 17 Claims
OG exemplary drawing
 
1. A semantic determining method, comprising:
acquiring, using a trained speech recognition model, a recognition character string matching with speech information;
acquiring, for each recognition character in the recognition character string, at least one entity vocabulary corresponding to the recognition character in an entity vocabulary library; and
determining, based on a hit of each entity vocabulary to the recognition character string, a matched entity vocabulary as a semantic determining result of the speech information,
wherein before acquiring the recognition character string matching with the speech information, the method further comprises:
establishing a descending index table matching with the entity vocabulary library, based on entity vocabularies in the entity vocabulary library, wherein the descending index table includes a plurality of key-value pairs, each of the key-value pairs includes a key name which is a character, and a value which is at least one entity vocabulary containing the character; and
the acquiring, for each recognition character in the recognition character string, at least one entity vocabulary corresponding to the recognition character in an entity vocabulary library comprising:
querying the descending index table matching with the entity vocabulary library, and acquiring the at least one entity vocabulary corresponding to the recognition character in the recognition character string.