Abstract: An automatic detection method for digital speech tamper by means of stitching was proposed. This method was based on speaking condition analyses, which comprised background noise analysis and speaker fettle analysis. Speech signals were divided into speech and silence segments. For silence segments containing noise, features in time domain and frequency domain were extracted for each frame. For each speech segment, rhythm and timbre features were extracted. Features changing points of the silence segments and speech segments were detected separately based on Bayesian information criterion, and the tamper detection result was obtained by integrative decision. The experimental results show that the proposed method can detect and locate the stitching points accurately.

Key words: digital speech, tamper detection, speaking condition analysis, background noise, prosodic characteristic, speech quality characteristic

摘要： 针对使用拼接手段的数字语音篡改,提出一种基于言语情境分析的篡改检测方法。该方法从背景噪声分析和说话人状态特征分析两方面入手,把语音信号分为语音部分和静音部分,对包含噪声的各个静音片段各帧提取时域和频域特征,对各语音片段提取韵律特征和音质特征,并分别基于贝叶斯信息准则检测特征的跳变点,通过综合判断得到篡改检测结果。实验结果表明,该方法能够比较准确地检测和定位语音拼接点。

关键词: 数字语音, 篡改检测, 言语情境分析, 背景噪声, 韵律特征, 音质特征