Toward standardizing perceptual voice quality measures

摘要 病理嗓音质量的感知在临床嗓音评估中是非常重要的，但足以 量化人的声音仍然是有问题的。2000年期间完成的研究数据 上一个资助期表明，与当前语音质量措施相关的许多困难来自 从定义和衡量质量的方式。我们建议发展一种心理声学 作为传统评级和声学分析协议的替代方案，这 心理声学模型将指定一组感知上重要的声学参数， 并由此量化语音的总体、整体质量。我们将首先确定声学的最小集合， 生成任何语音的合成副本所需的参数，使得收听者判断合成副本 与原始声音的质量相符。这一套将构成一个初步的心理声学模型的声音 质量.然后，我们将通过合成自然声音的副本来完善和验证这个心理声学模型 仅使用这些模型参数。在某种程度上，听众判断自然和合成标记 如果完全匹配，则心理声学模型将被认为是有效的。将对不匹配进行分析，以确定 应该在模型中增加或减少哪些参数。我们将评估 声学值的变化和语音偏离正常的程度的变化。这将提供 说明性模型，其指定声学参数如何在感知上联合收割机和交互以确定 沿着从“较好”到“较差”的连续体的任何声音样本的位置。“最后，我们将调查 在感知上重要的声学、频谱变化和声门的相关改变之间 配置.这种知识可以确定补救目标，这些目标最有可能 在治疗过程中产生声音改善。

ABSTRACT Perception of pathological voice quality is centrally important in clinical voice evaluation, but adequately quantifying the sound of a person's voice remains problematic. Data from studies completed during the previous funding period indicate that many difficulties associated with current measures of voice quality derive from the way in which quality is defined and measured. We propose the development of a psychoacoustic model of overall voice quality as an alternative to traditional ratings and acoustic analysis protocols. This psychoacoustic model will specify a set of perceptually-important acoustic parameters that combine to replicate and thereby quantify the overall, integral quality of a voice. We will first determine the minimal set of acoustic parameters required to produce a synthetic copy of any voice, such that listeners judge that the synthetic copy matches the quality of the original voice. This set will constitute a preliminary psychoacoustic model of voice quality. We will then refine and validate this psychoacoustic model by synthesizing copies of natural voices using only these model parameters. To the extent that listeners judge that the natural and synthetic tokens match exactly, the psychoacoustic model will be considered valid. Mismatches will be analyzed to determine what parameters should be added to or subtracted from the model. We will assess the relationship between changes in acoustic values and changes in the extent to which a voice deviates from normal. This will provide an explanatory model specifying how acoustic parameters combine and interact perceptually to determine the location of any voice sample along a continuum from "better" to "worse." Finally, we will investigate the link between perceptually-important acoustic, spectral changes and the associated alterations in glottal configuration. Such knowledge could identify targets for remediation that have the highest likelihood of producing vocal improvement during treatment.