基于常见伪装语音的声纹鉴定研究

Research on Voiceprint Authentication Based on Common Disguised Voices

摘要: 近年来电信网络诈骗案件呈现高发态势，声纹鉴定对此类案件的侦破具有重要作用。由于在近期电信网络诈骗案件当中较高频次出现的人为伪装语音，以及考虑到日常生活中人们佩戴口罩的频率较高情况。本文实验以一些日常生活中的实际案件里出现率较高的捏鼻伪装语音和戴口罩伪装语音两种人为语音伪装的方式为具体研究对象，采用听觉分析和语谱图分析两种不同分析相结合的方法，同时结合定量的分析结果，运用成对比较T检验方法对两种变异语音中体现发音人本身稳定性的一些典型的声学特征进行研究分析。本文的研究可以为语音识别技术的实际应用提供一定的依据和参考，为公安实战中基于变异语音案件的解决提供了一种新的思路。

Abstract: In recent years, telecom and internet fraud cases have shown a high incidence trend, where voiceprint identification plays a crucial role in solving such cases. Given the frequent occurrence of artificial voice impersonation in modern telecom fraud and considering the requirement for daily mask-wearing during COVID-19 prevention measures, this study focuses on two common voice imper-sonation methods—nose-pinch impersonation and mask-wearing impersonation—found in real-life cases. By combining auditory analysis with spectral graph analysis, and integrating quantitative results, we employ pairwise t-test to investigate typical acoustic features reflecting speaker stability in these variants. This research provides practical references for voice recognition technology applications and offers a novel approach for solving voice variation cases in police investigations.