首页|期刊导航|网络与信息安全学报|基于双流网络与校准增强的LPC语音隐写分析方法

基于双流网络与校准增强的LPC语音隐写分析方法OA

LPC speech steganalysis method based on dual-stream network and calibration enhancement

中文摘要英文摘要

网络语音流隐写分析是信息隐藏领域的研究热点.针对线性预测编码(linear predictive coding,LPC)语音隐写分析中内容与扰动强耦合、局部弱异常难表征,以及复杂场景下性能衰减等问题,提出一种融合双流网络与校准增强的LPC语音隐写分析方法.该方法构建原始流与校准流并行的双流框架,通过对比建模抑制语音内容干扰.校准流采用高斯统计扰动与重量化策略,以近似模拟语音编码中的重压缩效应.此外,设计了感知分离块解耦量化索引序列的帧间相关性与通道统计依赖,并引入多尺度注意力机制增强判别特征表达.性能评估结果表明,在低嵌入率与短样本等复杂场景下,所提方法的检测准确率优于现有主流方法,在检测准确率、虚警率与漏检率之间取得了更优平衡,展现出较强的综合检测性能与鲁棒性.

Network speech stream steganalysis is a research hotspot in the field of information hiding.To address the issues in linear predictive coding(LPC)-based speech steganalysis,such as strong coupling between content and perturbation,difficulty in characterizing local weak anomalies,and performance degradation in complex scenarios,proposes a dual-stream network with calibration enhancement for LPC speech steganalysis.The proposed method constructs a parallel dual-stream framework consisting of an original stream and a calibration stream,and sup-presses interference from speech content through contrastive modeling.The calibration stream adopts Gaussian sta-tistical perturbation and requantization strategies to approximately simulate the recompression effect in speech cod-ing.Furthermore,a perception separation block is designed to decouple inter-frame correlations and channel-wise statistical dependencies in the quantized index sequence,and a multi-scale attention mechanism is introduced to en-hance discriminative feature representation.The results of performance evaluation indicate that under complex sce-narios such as low embedding rates and short samples,the proposed method achieves higher detection accuracy than existing mainstream methods,strikes a better balance among detection accuracy,false alarm rate,and missed detection rate,and exhibits strong comprehensive detection performance and robustness.

王钟江;田晖;汤莉莉

华侨大学计算机科学与技术学院,福建 厦门 361021||厦门市数据安全与区块链技术重点实验室,福建 厦门 361021华侨大学计算机科学与技术学院,福建 厦门 361021||厦门市数据安全与区块链技术重点实验室,福建 厦门 361021厦门市数据安全与区块链技术重点实验室,福建 厦门 361021||华侨大学机电及自动化学院,福建 厦门 361021

信息技术与安全科学

网络语音流语音隐写分析线性预测编码双流网络多尺度特征建模

network speech streamspeech steganalysislinear predictive codingdual-stream networkmulti-scale feature modeling

《网络与信息安全学报》 2026 (4)

90-101,12

国家自然科学基金资助项目(No.62572202)福建省科技兴警科研专项项目(No.2025Y0068)The National Natural Science Foundation of China(No.62572202),Funds for the Innovation of Policing Sci-ence and Technology,Fujian province(No.2025Y0068)

10.11959/j.issn.2096-109x.AQ26056

评论