US 11,887,620 B2
Language model score calculation apparatus, language model generation apparatus, methods therefor, program, and recording medium
Ryo Masumura, Tokyo (JP); Tomohiro Tanaka, Tokyo (JP); and Takanobu Oba, Tokyo (JP)
Assigned to NIPPON TELEGRAPH AND TELEPHONE CORPORATION, Tokyo (JP)
Appl. No. 17/429,169
Filed by NIPPON TELEGRAPH AND TELEPHONE CORPORATION, Tokyo (JP)
PCT Filed Jan. 27, 2020, PCT No. PCT/JP2020/002650
§ 371(c)(1), (2) Date Aug. 6, 2021,
PCT Pub. No. WO2020/162240, PCT Pub. Date Aug. 13, 2020.
Claims priority of application No. 2019-021546 (JP), filed on Feb. 8, 2019.
Prior Publication US 2022/0013136 A1, Jan. 13, 2022
Int. Cl. G10L 25/30 (2013.01); G06N 3/08 (2023.01); G10L 25/51 (2013.01)
CPC G10L 25/30 (2013.01) [G06N 3/08 (2013.01); G10L 25/51 (2013.01)] 14 Claims
OG exemplary drawing
 
1. A language model score calculation device comprising a processor configured to execute operations comprising:
determining a history speech meta-information vector, wherein the history speech meta-information vector represents meta-information of a preceding speech using a meta-information understanding device regarding at least one piece of meta-information from a word string of the preceding speech;
converting the word string of the preceding speech and a speaker label representing a speaker of the preceding speech to a history speech embedding vector using a model parameter of a language model;
combining the history speech meta-information vector and the history speech embedding vector to obtain a speech unit combination vector;
converting a plurality of speech unit combination vectors obtained for past speech sequences to a speech sequence embedding vector using the model parameter of the language model; and
determining a language model score of a current speech from a word string of the current speech, a speaker label representing a speaker of the current speech, and the speech sequence embedding vector using the model parameter of the language model.