Speech detection method, speech detection apparatus, electronic device, and storage medium

CN116543794BActive Publication Date: 2026-07-07PING AN TECH (SHENZHEN) CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
PING AN TECH (SHENZHEN) CO LTD
Filing Date
2023-05-31
Publication Date
2026-07-07

AI Technical Summary

Technical Problem

Existing technologies often suffer from the loss of speech feature information when detecting whether speech data is synthesized, resulting in low accuracy of speech detection and an inability to effectively distinguish between synthesized and non-synthetic speech.

Method used

A feature extraction network, first and second attention networks are used to reconstruct multi-scale features from speech data. A detection network is then used to score the speech and generate a target speech feature map to improve the accuracy and comprehensiveness of feature information.

Benefits of technology

By employing multi-scale feature reconstruction and scoring mechanisms, the accuracy of speech detection is significantly improved, enabling better recognition of synthetic speech and enhancing the security and reliability of online transactions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116543794B_ABST
    Figure CN116543794B_ABST
Patent Text Reader

Abstract

The application provides a speech detection method, a speech detection device, an electronic equipment and a storage medium, and belongs to the technical field of financial technology. The method comprises the following steps: acquiring original speech data; inputting the original speech data into a speech detection model; performing feature extraction on the original speech data based on a feature extraction network to obtain an original speech feature map; performing first multi-scale feature reconstruction on the original speech feature map based on a first attention network to obtain an initial speech feature map; performing second multi-scale feature reconstruction on the initial speech feature map based on a second attention network to obtain a target speech feature map; performing speech scoring on the target speech feature map based on a detection network to obtain a target speech score; performing synthetic speech detection according to the target speech score to obtain speech detection data; and the speech detection data is used for representing whether the original speech data is synthetic speech or non-synthetic speech. The application can improve the accuracy of speech detection.
Need to check novelty before this filing date? Find Prior Art