Title:
SPEECH DECODING METHOD AND APPARATUS, COMPUTER DEVICE AND STORAGE MEDIUM
Document Type and Number:
WIPO Patent Application WO/2020/119351
Kind Code:
A1
Abstract:
Disclosed is a speech decoding method, which is implemented by a computer device, wherein speech comprises a current audio frame and a previous audio frame. The method comprises: acquiring, from a first token list, a target token corresponding to the minimum decoding score (403), wherein the first token list comprises first tokens which are obtained by decoding a previous audio frame in different decoding networks, each of the first tokens comprises a state pair and a decoding score, and the state pair is used for representing a correlation between a first state of the first token in a first decoding network corresponding to a low-order language model and a second state thereof in a second decoding network corresponding to a differential language model; according to the target token and an acoustic vector of a current audio frame, determining a pruning parameter when the current audio frame is decoded (404); and according to the first token list, the pruning parameter and the acoustic vector, decoding the current audio frame (405).
Inventors:
HUANG YIHENG (CN)
JIAN XIAOZHENG (CN)
HE LIQIANG (CN)
JIAN XIAOZHENG (CN)
HE LIQIANG (CN)
Application Number:
PCT/CN2019/116686
Publication Date:
June 18, 2020
Filing Date:
November 08, 2019
Export Citation:
Assignee:
TENCENT TECH SHENZHEN CO LTD (CN)
International Classes:
G10L15/193; G10L15/18
Foreign References:
CN110164421A | 2019-08-23 | |||
CN108305634A | 2018-07-20 | |||
CN108288467A | 2018-07-17 | |||
CN105513589A | 2016-04-20 | |||
CN105575386A | 2016-05-11 | |||
CN105654945A | 2016-06-08 | |||
CN1455387A | 2003-11-12 | |||
CN108682415A | 2018-10-19 |
Attorney, Agent or Firm:
ADVANCE CHINA IP LAW OFFICE (CN)
Download PDF: