Title:
推論器、推論プログラムおよび学習方法
Document Type and Number:
Japanese Patent JP7385900
Kind Code:
B2
Abstract:
To provide a multilingual end-to-end voice recognition system by using a model of a smaller parameter size.SOLUTION: An inference unit for receiving input of a voice signal uttered in arbitrary language in a plurality of languages and outputting a corresponding text is provided. The inference unit comprises: a learned model which receives an input sequence indicating a voice feature of the voice signal and outputs expression of a level different from a character level, which indicates the feature of a character included in the corresponding text; and a re-constitution section for re-constituting the corresponding text from expression outputted from the learned model by referring to a corresponding relation between a previously determined character and the feature of the character.SELECTED DRAWING: Figure 3
More Like This:
Inventors:
Lee Sun
Rushugan
Ding Jinchen
Tatsuya Kawahara
Hisashi Kawai
Rushugan
Ding Jinchen
Tatsuya Kawahara
Hisashi Kawai
Application Number:
JP2019163555A
Publication Date:
November 24, 2023
Filing Date:
September 09, 2019
Export Citation:
Assignee:
National Institute of Information and Communications Technology
International Classes:
G10L15/10; G10L15/16
Domestic Patent References:
| JP9500223A | ||||
| JP2010072446A | ||||
| JP2014229124A |
Foreign References:
| US5758023 | ||||
| US20160110642 | ||||
| US20190189111 |
Attorney, Agent or Firm:
Patent Attorney Fukami Patent Office