文章摘要
肖东,莫福源,陈庚,马力.低码率语音编码中过渡帧对合成语音的影响*[J].,2016,35(1):77-83
低码率语音编码中过渡帧对合成语音的影响*
Effect of transition frame on synthesized speech in low bit rate speech coding
投稿时间:2015-06-09  修订日期:2015-12-22
中文摘要:
      过渡段对语音清晰度、可懂度和人耳听觉感知都起到不可忽视的作用。参数语音编码中,包含有过渡段的语音帧能否得到恰当处理,是决定其合成语音是否清晰可懂的关键。本文以混合激励线性预测编码为参考,将其中的语音帧划分为静音、清音、浊音、过渡四大类后分别处理,在以往低码率语音编码(<1kbps)工作基础上,比较了八种过渡帧划分方法对合成语音PESQ MOS的影响。经分析后发现:不同的过渡帧对PESQ MOS的贡献也不同。由清、静音向浊音变化的过渡帧的贡献最大;介于浊辅音与元音之间的过渡帧的贡献也不应被忽略。
英文摘要:
      Transition segments play an essential role in clarity, intelligibility and auditory perception of speech. In parametric speech codec algorithm, whether the synthesized speech is clear and intelligible is critically determined by whether transition frames, which contain the transition segments, can be processed felicitously. Referring to MELP (Mixed Excitation Linear Prediction), frames are classified into four types: silent, unvoiced, voiced and transition. Each type is processed respectively. Based on the previous work of low bit rate (<1kbps) speech coding, the effect of 8 transition frame classification methods on PESQ MOS (Perceptual Evaluation of Speech Quality Mean Opinion Score) are studied. It is found that: different transition contributes differently to PESQ MOS. The transition from unvoiced or silent frame to voiced frame is the most important. And the transition between voiced consonant and vowel can not be neglected either.
DOI:10.11684/j.issn.1000-310X.2016.01.011
中文关键词: 低码率语音编码,混合激励线性预测编码,过渡段
英文关键词: Low bit rate speech codec,MELP,Transition segment
基金项目:(61302109);
作者单位E-mail
肖东* 中国科学院声学研究所水声环境特性实验室 北京 #$NL 中国科学院声学研究所 北京 xiaodong@mail.ioa.ac.cn 
莫福源 中国科学院声学研究所 mofuyuan@aliyun.com 
陈庚 中国科学院声学研究所水声环境特性实验室 gengchen333@gmail.com 
马力 中国科学院声学研究所水声环境特性实验室 mary1968@tom.com 
摘要点击次数: 2234
全文下载次数: 1709
查看全文   查看/发表评论  下载PDF阅读器
关闭