Title:
ERROR CORRECTION DEVICE, ERROR CORRECTION METHOD, AND PROGRAM
Document Type and Number:
WIPO Patent Application WO/2022/162767
Kind Code:
A1
Abstract:
The present invention provides a high-accuracy error correction technology of a speech recognition result. This error correction device comprises: a first distributed representation sequence generation unit that generates, from a first token sequence representing a speech recognition result, a first distributed representation sequence being a sequence of distributed representations of tokens being elements of the first token sequence; a second distributed representation sequence generation unit that generates, from an input speech-related data sequence being a sequence of acoustic feature amounts of speech or vectors generated from the speech, a second distributed representation sequence being a sequence of distributed representations of input speech-related data being elements of the input speech-related data sequence; a distributed representation sequence integration unit that generates an integrated distributed representation sequence being a sequence of distributed representations including distributed representations being elements of the first distributed representation sequence and the second distributed representation sequence; an encoding unit that generates, from the integrated distributed representation sequence, an encoded integrated distributed representation sequence being a sequence of distributed representations corresponding to the input speech-related data and the feature of the speech recognition result; and a decoding unit that generates, from the encoded integrated distributed representation sequence, a second token sequence representing an error correction result of the speech recognition result.
Inventors:
TANAKA TOMOHIRO (JP)
MASUMURA RYO (JP)
MASUMURA RYO (JP)
Application Number:
PCT/JP2021/002761
Publication Date:
August 04, 2022
Filing Date:
January 27, 2021
Export Citation:
Assignee:
NIPPON TELEGRAPH & TELEPHONE (JP)
International Classes:
G10L15/16; G10L15/22
Domestic Patent References:
| WO2019163718A1 | 2019-08-29 |
Foreign References:
| JP2020527758A | 2020-09-10 | |||
| JP2019139010A | 2019-08-22 | |||
| JP2010134074A | 2010-06-17 | |||
| US20180330730A1 | 2018-11-15 | |||
| US20190189115A1 | 2019-06-20 |
Other References:
HU, KE ET AL.: "DELIBERATION MODEL BASED TWO-PASS END-TO-END SPEECH RECOGNITION", ICASSP, vol. 20, 20 May 2020 (2020-05-20), pages 7799 - 7803, XP033793266, DOI: 10.1109/ICASSP40776.2020.9053606
Attorney, Agent or Firm:
NAKAO, Naoki et al. (JP)
Download PDF: