[1]魏春雨,孙 蒙,贾 冲.说话人重识别中的基频和共振峰联合还原方法[J].计算机技术与发展,2023,33(06):47-53.[doi:10. 3969 / j. issn. 1673-629X. 2023. 06. 008]
 WEI Chun-yu,SUN Meng,JIA Chong.Joint Restoration of Pitch and Formant for Speaker Re-recognition[J].,2023,33(06):47-53.[doi:10. 3969 / j. issn. 1673-629X. 2023. 06. 008]
点击复制

说话人重识别中的基频和共振峰联合还原方法()
分享到:

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
33
期数:
2023年06期
页码:
47-53
栏目:
媒体计算
出版日期:
2023-06-10

文章信息/Info

Title:
Joint Restoration of Pitch and Formant for Speaker Re-recognition
文章编号:
1673-629X(2023)06-0047-07
作者:
魏春雨孙 蒙贾 冲
陆军工程大学 指挥控制工程学院,江苏 南京 210007
Author(s):
WEI Chun-yuSUN MengJIA Chong
School of Command and Control Engineering,Army Engineering University of PLA,Nanjing 210007,China
关键词:
说话人匿名说话人重识别McAdams 系数共振峰还原
Keywords:
speaker anonymizationspeaker re-recognitionMcAdams coefficientformantrestoration
分类号:
TP391. 9
DOI:
10. 3969 / j. issn. 1673-629X. 2023. 06. 008
摘要:
话人匿名技术的出现,对基于声纹的生物特征识别造成了巨大的安全威胁。 对于利用各种变声工具实施的说话人匿名,匿名语音中的说话人个性特征相比原始语音发生了显著改变,会严重影响说
话人识别的效果。 针对现有说话人重识别方法存在的语音还原手段单一、在变声工具类型未知情况下的匿名语音还原效果尚不明确等问题,提出了一种基于基频和共振峰联合还原的说话人变声匿名重识别方法。 该方法在基频逆变换变声还原的基础上,引入 McAdams 系数调整语音的共振峰,同时使用基于 x-vector 的说话人识别模型进行声纹相似度评分,提高了黑盒变声匿名条件下还
原语音与真实语音的声学特征相似度,增强了说话人识别系统对不同类型变声匿名语音的重识别能力。 实验结果表明,提出的方法对四种音频编辑软件和三种真实变声器材匿名语音的重识别效果均优于现有基线重识别方法。
Abstract:
The emergence of speaker anonymization poses a huge security threat to biometric recognition based on voiceprint. For thespeaker anonymization implemented by various voice changing tools,the personality characteristics of the speaker in the anonymous voicehave changed significantly compared with the original voice,which will seriously affect the effect of speaker recognition. Aiming at theproblems of single speech restoration means and unclear effect of anonymous speech restoration in the case of unknown voice modificationtools in existing speaker re - recognition methods,a speaker re - recognition method of anonymous voices joint restoration of pitch andformant based is proposed. Besides the pitch inversion transformation, the proposed method introduces the McAdams coefficient tomodify the formant characteristics of voices,and uses the speaker recognition model based on x-vector to calculate the utterance-levelsimilarity score,which improves the acoustic similarity between the restored voice and the real voice under the condition of black boxvoice changing,and enhances the ability of speaker recognition system to recognize different kinds of anonymous voices. Experimentalresults show that the proposed method has better performance than the existing baseline method in restoring the anonymous voicesgenerated by four audio editing software and three physical voice changing tools.
更新日期/Last Update: 2023-06-10