一种用于识别非法广播的音频多特征比对方法

An Audio Comparison Method Based on Multi-features for Illegal Broadcast Detect

部分非法广播占用正常广播的频点进行播放,极具隐蔽性。为识别此类非法广播,需要采用更加综合化和智能化的手段。本文致力于通过广播音频在信号层面和语义层面特征的融合分析方法来检测识别,具体措施是:在信号层面,通过计算短时能量、短时过零率、频谱质心等指标来反映广播的信号特征;在语义层面,通过在语音识别后进行文本词频统计进而形成关键词序列来反映广播的内容特征;最后,建立基于信号和语义的多特征判决规则来进行多级融合判决,从而识别是否为存在占用正常广播频点的非法广播。基于仿真数据和真实数据的实验表明了本文方法不仅具有有效性,而且具有工程适用价值。本文方法有助于确保我国广播信息空间安全。

Some illegal FM broadcasts occupy frequency points of normal broadcasts, which is very covert. In order to detect such kind of illegal broadcasts, more comprehensive and intelligent measures should be taken. This paper focuses on detecting illegal broadcast based on both signal features and semantic features. Firstly, at signal level, short-term energy, short-term zero-crossing rate, spectral centroid and other indicators are extracted to refect signal characteristics;secondly, at semantic level, text word frequency statistics based on speech recognition is conducted to present semantic characteristics;finally, a decision rule based on multi-features fusion is established to detect illegal broadcast occupying frequency points of normal broadcast. Experiments based on simulation and practical data show the proposed method is not only effective, but also has engineering application value. This paper offers a reference for ensuring broadcast security.