Measuring Breathiness with an Auditory Processing Model

描述（由申请人提供）：语音质量的测量对于做出有关语音障碍患者管理的若干决策很重要。研究表明，在常规临床设置中对语音质量的主观评分是高度不一致的，可能不是有效的语音质量衡量标准。另一种测量语音质量的方法是开发与听者对语音质量的感知密切相关的声学测量。然而，到目前为止，没有一种测量方法显示出与听者对所研究的音质的感知有高度和一致的相关性。声学测量与语音质量的感知评级之间的相关性差和不一致可能源于声学信号与其感知相关性之间的非线性和多维关系。物理刺激与其心理物理关联之间的非线性关系在所有人类感觉系统中都可以看到。例如，在听觉领域，强度和响度之间的关系可以用幂律来更好地解释，频率和音调之间的关系可以用梅尔或巴克尺度来更好地映射。心理声学和听觉生理学的最新进展导致了几种听觉模型的发展，这些模型能够解释这种非线性现象。因为这些非线性过程是所有听觉感知任务所固有的，它们也会影响特定的声音信号如何影响语音质量的感知。本研究将探讨使用听觉处理模型作为信号处理前端来研究语音质量。假设这种听觉模型的输出将提供语音质量的客观测量，显示出与语音质量的感知评级高度一致的相关性。本研究旨在(1)研究声音信号的特定变化对呼吸语音质量感知的影响，(2)使用听觉处理模型来了解听觉频谱的变化导致感知语音质量的变化，以及(3)制定客观措施来解释合成元音和自然元音的感知数据。

DESCRIPTION (provided by applicant): Measurement of voice quality is important for making several decisions regarding the management of patients with voice disorders. It has been shown that subjective ratings of voice quality made in a routine clinical set-up are highly inconsistent, and may not be valid measures of voice quality. An alternative approach to measure voice quality is to develop acoustic measures that closely correspond with listeners' perception of voice quality. However, none of the measures developed thus far show a high and consistent correlation with listener's perception of the vocal quality under study. The poor and inconsistent correlation between acoustic measures and perceptual ratings of voice quality may arise from the non-linear and multidimensional relationship between the acoustic signal and its perceptual correlate. Non-linear relationships between a physical stimulus and its psychophysical correlate are seen in all human sensory systems. For example, in the auditory domain the relation between intensity and loudness is better explained by the power law and that between frequency and pitch is better mapped using the Mel or the Bark scale. Recent advances in psychoacoustics and auditory physiology has led to the development of several auditory models which are capable of accounting for such non-linear phenomena. Because these non-linear processes are inherent to all auditory perceptual tasks, these also affect how a specific vocal acoustic signal affects the perception of voice quality. This research will investigate the use of an auditory processing model as a signal processing front end to study voice quality. It is hypothesized that the output of such an auditory model will provide objective measures of voice quality that show a high and consistent correlation with perceptual ratings of voice quality. The proposed research seeks to (1) study the effects of specific changes in vocal acoustic signal on the perception of breathy voice quality, (2) use an auditory processing model to understand what changes in the auditory spectrum lead to the changes in perceived voice quality, and (3) develop objective measures to explain perceptual data in synthetic and natural vowels.