基于倒谱分析的实时广播音频相似度快速比对算法
Fast Similarity Comparison Algorithm for Real-time Broadcast Audio Based on Cepstrum Analysis
作者：邵玉斌(昆明理工大学 信息工程与自动化学院, 云南 昆明 650500)；唐传林(昆明理工大学 信息工程与自动化学院, 云南 昆明 650500)；赵至柔(昆明理工大学 信息工程与自动化学院, 云南 昆明 650500)；龙华(昆明理工大学 信息工程与自动化学院, 云南 昆明 650500)；杜庆治(昆明理工大学 信息工程与自动化学院, 云南 昆明 650500)
Author：SHAO Yubin(School of Info. Eng. and Automation, Kunming Univ. of Sci. and Technol., Kunming 650500, China)；TANG Chuanlin(School of Info. Eng. and Automation, Kunming Univ. of Sci. and Technol., Kunming 650500, China)；ZHAO Zhirou(School of Info. Eng. and Automation, Kunming Univ. of Sci. and Technol., Kunming 650500, China)；LONG Hua(School of Info. Eng. and Automation, Kunming Univ. of Sci. and Technol., Kunming 650500, China)；DU Qingzhi(School of Info. Eng. and Automation, Kunming Univ. of Sci. and Technol., Kunming 650500, China)
收稿日期：2019-08-08 年卷（期）页码：2020,52(3):178-185
期刊名称：工程科学与技术
Journal Name：Advanced Engineering Sciences
关键字：音频比对;延时估计;倒谱分析;实时广播音频
Key words：audio comparison;delay estimation;cepstrum analysis;real-time broadcast audio
基金项目：国家自然科学基金地区科学基金项目（61761025）
中文摘要
为了解决广播音频中经常存在噪声干扰和时间延迟导致音频比对结果不准确的问题，提出具有延时自适应意识的音频比对算法。针对常用算法中测量音频特征距离抗噪性能差的不足，采用倒谱对两音频的混合信号分析，并利用倒谱对功率谱中的等距离频率成分有很强的分辨能力这一特性来进行自适应延时估计和比对；为比对不同情况的两音频都可得到准确的相似度，提出对其中一音频加入短延时，再将两音频叠加混合后做倒谱分析；并根据加入不同短延时的效果选择出最优短延时，进一步提升算法性能。使用真实广播不同节目中截取出来的多个音频，在无噪声和不同信噪比加性高斯白噪声条件下，通过仿真实验评估了所提出算法的性能，比较了不同信噪比下的延时估计结果和音频相似度。实验结果证明，所提出方法的延时估计结果和比对结果优于现有算法，在低信噪比（SNR=2 dB）下，也可以达到90.36%的音频比对匹配精度，且计算速度能够达到实时比对的要求。
英文摘要
In order to solve the problem of inaccurate audio comparison caused by noise and delay in broadcast audio, an audio comparison algorithm with delay-adaptive-aware was proposed. To tackle the poor noise immunity of audio feature distance measurement in conventional algorithms, cepstrum was used to analyze the mixed signal of two audio to estimate delay adaptively and comparison, which has a strong ability to resolve equidistant frequency components in the power spectrum. Then, a method of short delay was proposed to obtain accurate similarity between two audios in different situations, which is added short delay in one of the audio. Afterwards, the optimal short delay was selected according to the effect of adding different short delay, so as to improve the performance of the algorithm further. Finally, the simulation experiments were conducted to evaluate the performance of the proposed algorithm, in which multiple audio clips of different broadcast programs were utilized under the condition of different SNR and additive white Gaussian noise. And the delay estimation results and audio similarity under different SNR are compared to verify the effectiveness of the algorithm. Experimental results show that the proposed algorithm outperforms existing algorithms and can achieve 90.36% audio comparison matching accuracy at a low SNR (2 dB), and the calculation speed can meet the requirement of real-time comparison.
【关闭】
