Title:
CROWD-INFORMATION-FUSED SPEECH EMOTION RECOGNITION METHOD AND SYSTEM
Document Type and Number:
WIPO Patent Application WO/2022/199215
Kind Code:
A1
Abstract:
A crowd-information-fused speech emotion recognition method and system, relating to the field of artificial intelligence. The method comprises the following steps: collecting a speech signal of a user (S1); preprocessing the speech signal to obtain a Mel spectrum (S2); removing mute segments before and after the Mel spectrum (S3); acquiring deep crowd information by means of a crowd classification network (S4); acquiring Mel spectrum depth information by means of a Mel spectrum preprocessing network (S5); acquiring fusion information by means of SENet fusion features (S6); and obtaining an emotion recognition structure by means of the classification network (S7). By fusing crowd information features, emotion feature extraction is more accurate, and by performing information fusion by means of a SENet channel attention mechanism, deep feature extraction can be effectively performed, and the overall recognition precision is improved.
Inventors:
LI TAIHAO (CN)
ZHENG SHUKAI (CN)
LIU YULONG (CN)
PEI GUANXIONG (CN)
MA SHIJIE (CN)
ZHENG SHUKAI (CN)
LIU YULONG (CN)
PEI GUANXIONG (CN)
MA SHIJIE (CN)
Application Number:
PCT/CN2022/070728
Publication Date:
September 29, 2022
Filing Date:
January 07, 2022
Export Citation:
Assignee:
ZHEJIANG LAB (CN)
International Classes:
G10L25/63; G10L25/24; G10L25/30
Foreign References:
| CN112712824A | 2021-04-27 | |||
| CN110021308A | 2019-07-16 | |||
| CN110491416A | 2019-11-22 | |||
| US6173260B1 | 2001-01-09 | |||
| CN109935243A | 2019-06-25 | |||
| CN110164476A | 2019-08-23 | |||
| US10937446B1 | 2021-03-02 |
Attorney, Agent or Firm:
HANGZHOU ZHEKE PATENT AGENCY (GENERAL PARTNERSHIP) (CN)
Download PDF: