MODEL TRAINING METHOD, VOICE DETECTION AND LOCALIZATION METHOD, APPARATUS, DEVICE, AND MEDIUM

WIPO Patent Application WO/2023/273469

A1

A model training method, a voice detection and localization method, an apparatus, a device, and a medium. The method comprises: receiving speech content by means of a microphone array, and obtaining multi-channel audio, the speech content being voice speech content or noise speech content; obtaining a target vector and an audio feature parameter of the multi-channel audio, the target vector comprising N labels, the N labels being in one-to-one correspondence with N spatial regions, and each label indicating the probability that the corresponding spatial region contains a voice; taking the audio feature parameter of the multi-channel audio to serve as training sample input, taking the target vector of the multi-channel audio to serve as training sample target output, utilizing a training sample to train a deep neural network, and obtaining a target model. The accuracy of the target model is thereby increased, and the accuracy of voice detection and localization is consequently increased.

JP4306424	Voice section detector
JPS61203497	VOICE RECOGNITION SYSTEM
JPS6057396	CALCULATION OF THRESHOLD

HAN KEWEI (CN)

PCT/CN2022/084599

January 05, 2023

March 31, 2022

Click for automatic bibliography generation  

CLOUDMINDS ROBOTICS CO LTD (CN)

G10L25/78; G01S3/802

WO2021013346A1

2021-01-28

CN111142066A	2020-05-12
CN112799016A	2021-05-14
CN105068048A	2015-11-18
CN110794368A	2020-02-14
US20170040030A1	2017-02-09

TEKYRS INTELLECTUAL PROPERTY INC. (CN)

View/Download PDF      PDF Help

Previous Patent: COMMUNICATION METHOD, APPARATUS AND SYSTEM

Next Patent: DISPLAY PANEL, DISPLAY SCREEN AND ELECTRONIC DEVICE