合成语音与自然语音嗓音的声学对比分析——以18种语言为例  

Acoustic Comparison Analysis between Synthesized and Natural Speech Voices:An Example Using 18 Languages

作　　者：冉启斌[1] 黄玮 RAN Qibin;HUANG Wei

机构地区：[1]南开大学文学院

出　　处：《天津外国语大学学报》2024年第5期73-87,112,F0003,共17页Journal of Tianjin Foreign Studies University

基　　金：国家社会科学基金重大项目“中国境内语言核心词汇声学数据库及计算研究”(19ZDA300)。

摘　　要：文章以18种语言的合成语音和自然语音为语料,考察了合成语音与自然语音在基频微扰、振幅微扰和谐噪比三个嗓音参数上的差异。实验表明:18种语言的合成语音基频微扰均要大于自然语音,其中15种具有统计上的显著差异;14种语言的合成语音振幅微扰大于自然语音,其中13种具有统计上的显著差异;17种语言的合成语音谐噪比小于自然语音,其中15种具有统计上的显著差异。合成语音基频微扰、振幅微扰和谐噪比之间的相关性均比自然语音小。合成语音在声带振动的频率、振幅和嗓音信号的周期性上倾向于具有更强的不规律性。This paper examines the differences between synthesized and natural speech in terms of three acoustic parameters:jitter,shimmer,and harmonic-to-noise ratio(HNR),using data from 18 languages.The experiment shows that the jitter of synthesized speech is larger than that of natural speech in all 18 languages,with statistically significant differences in 15 of them.The shimmer of synthesized speech is larger than that of natural speech in 14 languages,with statistically significant differences in 13 of them.The HNR of synthesized speech is smaller than that of natural speech in 17 languages,with statistically significant differences in 15 of them.The correlations between jitter,shimmer,and HNR are smaller for synthesized speech than for natural speech.Synthesized speech tends to have greater irregularity in the frequency,amplitude,and periodicity of vocal cord vibration.

关 键 词：合成语音 嗓音声学分析 基频微扰 振幅微扰 谐噪比 

分 类 号：H01[语言文字—语言学]