摘要
为了直观地给听力语言障碍者提供发音改进信息,本文研究了利用不送气塞音时域和频域的声学特征来进行特征提取的算法,并运用到辅音训练辅助系统中。系统采用语音信号处理的方法,显示由线性预测得到的时频分布语谱图,用共振峰跟踪技术追踪不送气塞音与后接元音之间过渡音征的共振峰轨迹。由时域特征与共振峰轨迹,判别发音是否准确,并用视觉反馈的方式提供给受训者发音改进意见。系统对10人(6男,4女)的三个不送气塞音/b/、/d/、/g/加上元音/a/的发音进行了测试,获得了良好的初步结果。
In this paper, an algorithm of consonant feature extraction for the speech training aid is proposed. The voiced stops have several effective acoustic features achieved from time domain and frequency domain by speech processing, which can guide hearing impaired to practice articulation. The algorithm analyzes the three voice stops/b/, /d/ and/g/ with the successive vowel/a/by show-time linear predictive coding （LPC） , and then gives the spectrogram from which the first three formants of the whole speech can be tracked. The second and third formant trajectory of the transition cue between voiced stop and vowel give the information about place of articulation which can not only gives advice to inarticulate trainees but also identify the three voiced stop. A group of 10 subjects （6 males and 4 females）was selected to take part in a test of three voiced stops articulation, and a satisfactory result was gained.
出处
《北京生物医学工程》
2007年第3期292-296,共5页
Beijing Biomedical Engineering
关键词
聋哑康复
语音训练
辅音
线性预测
共振峰轨迹
hearing impaired rehabilitation
speech training
consonant
linear predictive coding （LPC）
formant trajectory
