音频数据的处理方法、装置、设备以及计算机可读介质
By using Fourier transform and feature data cancellation processing, the problem of noise superimposed on effective audio segments that cannot be removed in existing technologies is solved, thereby improving audio data quality and recognition accuracy.
CN116343782BActive Publication Date: 2026-07-17HEFEI IFLY DIGITAL TECH CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEFEI IFLY DIGITAL TECH CO LTD
- Filing Date
- 2023-03-30
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing noise reduction methods cannot effectively remove additive and multiplicative noise superimposed on valid audio segments, resulting in insufficient audio data quality and intelligibility, which affects the accuracy of audio recognition technology.
Method used
The effective and invalid tone audio data are converted from the time domain to the frequency domain by Fourier transform to obtain their feature data. The invalid tone feature data is then used to cancel out the feature data of the effective tone to remove the feature information of the invalid tone.
Benefits of technology
It improves the quality and intelligibility of audio data, enhances the recognition accuracy of audio recognition models, and reduces noise interference.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116343782B_ABST
Abstract
本申请提供一种音频数据的处理方法、装置、设备以及计算机可读介质，该方法通过获取原始音频数据的有效音音频数据和无效音音频数据；其中，有效音音频数据为原始音频数据中能够被识别的音频数据；无效音音频数据为原始音频数据中不能够被识别的音频数据；通过傅里叶变换将有效音音频数据由时间域转换为频率域，得到原始音频数据的有效音特征数据；并通过傅里叶变换将无效音音频数据由时间域转换为频率域，得到原始音频数据的无效音特征数据；使用原始音频数据的无效音特征数据对原始音频数据的有效音特征数据进行抵消处理，得到处理后的有效音特征数据，处理后的有效音特征数据不具有无效音的特征信息，提高了有效音的质量。
Need to check novelty before this filing date? Find Prior Art