呼吸声分类技术研究及检测系统设计*

患有呼吸问题的病人由于呼吸道腺体不受意识控制、异物或者其本身肺部病变等原因会出现呼吸异常。医护人员通过病人的呼吸状况可以找寻和分析病人出现这种呼吸状况的原因。而以往呼吸声的诊断方式是专业的医护人员通过使用听诊器对病人进行肺部听诊。但是听诊的结果取决于医护人员的经验与相关参数。在初级诊断治疗阶段,初级医生识别听诊声的效率以及准确率很低,一般从20%到80%不等。因此会存在10%到20%的高误诊率（漏诊、错诊和延误）。本文提出了一种利用PVDF薄膜传感器提取语声特征的检测系统,该检测系统根据病人发出声音的不同,提取呼吸声特征值判断病人的呼吸状况。通过PVDF传感器采集的微弱呼吸声再经过KNN算法分类之后其识别率可达90.6%,对于细分种较类多的湿罗声,其识别率在80.2%左右。综上表明相比于传统听诊方式通过PVDF传感器采集识别的结果具有更高的准确性和可靠性,其判断呼吸状况的结果可以为医护人员提供参考,更好的为患有呼吸疾病的病人提供监测。     

Synthesis of voiced sounds using low-dimensional models of the vocal cords and time-varying subglottal pressure

The vocal cords play an important role on voice production. Air coming from the lungs is forced through the narrow space between the two vocal cords that are set in motion in a frequency that is governed by the tension of the attached muscles. The motion of the vocal cords changes the type of flow, that comes from the lungs, to pulses of air, and as the flow passes through the oral and nasal cavities, it is amplified and further modified until it is radiated from the mouth. This complex process can be modeled by a system of integral-differential equations. This paper considers two mechanical models previously used for explaining the dynamics of the vocal cords. It shows that the level of naturalness of the sound generated by these models is rather poor, and it proposes temporal variations of the parameters of the models to increase such level. Examples of synthetic vowels and diphthongs are given to assess the models. In general, the results show that, although the system of voice production is complex, we can achieve satisfactory results with relatively simple low-dimensional models, by suitable temporal variations of the aerodynamic parameters.