摘要
梅尔倒频谱系数(MFCC)是语音识别、说话人识别领域的常用技术,近年来也被应用于音频内容的一致性比对领域。本文提出了一种基于MFCC帧内二值差分的音频指纹提取算法,并重点设计了动态时延自适应的特征比对机制,经过实验环境模拟测试及在播出系统中的部署实测,本算法可有效适应不同类型的音频节目比对,检测精度高、时延自适应性能力强,有效地提升了安全播出保障能力,具有较高的推广价值。
Mel-Frequency Cepstral Coefficients(MFCC) is often used in automatic speech recognition and speaker recognition,and is also applied to consistency comparison of audio content in recent years.This paper proposes an audio fingerprint extraction algorithm based on MFCC intra-frame binary differential,and designs an adaptive feature matching mechanism in dynamic time-delay situation.According to the tests and application in both experimental environment and on-line broadcasting system,it can be concluded that the algorithm is adaptable,highly precise,and effective for comparison of different types of audio programs,which can truly promote the ability of safe broadcasting and has great application value.
作者
张杨
Zhang Yang(China Media Group,Beijing 100866,China)
出处
《广播与电视技术》
2022年第6期128-134,共7页
Radio & TV Broadcast Engineering
关键词
音频比对
MFCC
二值差分
时延自适应
Audio comparison
MFCC
Binary differential
Time-delay adaptive
