皮肤听声原理在语音合成中的应用研究  

Research on application of skin-hearing theory in speech synthesis

作　　者：李建文[1] 朱悦 LI Jianwen;ZHU Yue(School of Electronic Information and Artificial Intelligence,Shaanxi University of Science and Technology,Xi’an 710021,China)

机构地区：[1]陕西科技大学电子信息与人工智能学院,陕西西安710021

出　　处：《现代电子技术》2020年第19期35-39,44,共6页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(60672001)。

摘　　要：语音合成一直是信息交互的重要研究领域,但是目前语音合成的方法还远不够完备。为提高重建语音的辨识正确率,提出以频谱构造法进行语音信号合成。首先将语音信号经过去噪、加窗、分帧和傅里叶变换等处理得到语音频谱图,然后通过频率解析提取共振峰谱线关键频率信息,在以C#搭建的智能语音合成平台上进行语音信号重建,最后利用重建语音信号和原始标准语音信号进行主观辨析测试。实验结果表明,重建语音信号可平衡各频率段的能量,突出语音信号的频谱特征。相比双谱线汉语重建语音,除汉语音素[o],其他单韵母音素识别正确率皆有明显提高。The speech synthesis has always been an important research area of information interaction,however,the current methods of speech synthesis are far from complete.In order to improve the recognition accuracy of reconstructed speech,a speech signal synthesis method is proposed based on spectrum construction.The speech signal is processed by denoising,windowing,framing and Fourier transform to obtain a speech spectrum diagram,and then the key frequency information of the formant spectral line is extracted by frequency analysis.Speech signal is reconstructed on the intelligent speech synthesis platform build with C#.In the end,the reconstructed speech signal and original standard speech signal are used in the subjective discrimination analysis test.The experimental results show that the reconstructed speech signal can balance the energy of each frequency band and highlight the spectrum feature of speech signal.In comparison with the Chinese reconstructed speech with double-spectrum line,the accuracy of single final phonemes is significantly improved except for the Chinese phoneme[o].

关 键 词：皮肤听声 语音信号处理 语音频谱图 频谱特征 共振峰谱线 频谱构造 语音信号重建 浊音 

分 类 号：TN912-34[电子电信—通信与信息系统]