Towards Standardizing Perceptual Voice Quality Measures

语音质量的判断在很大程度上影响了患者对自己声音的看法，也影响了临床医生的决策

Judgments of voice quality contribute greatly to patients' opinions of their voice, and to a clinician's decision to initiate and continue treatment. Quality judgments are also used as a standard against which instrumental measures of voice are validated. Nevertheless, these "subjective" measures of voice quality are not highly regarded as either clinical or research tools, because of problems with reliability and untested validity. We hypothesize that problems in voice quality measurement do not originate within the listener, but rather derive from the methods used to measure what listeners hear. The proposed research departs from traditional rating scale methods by applying a speech synthesizer for pathological voice quality to study basic issues concerning reliability and validity of voice quality measures, and to determine the perceptual importance of the acoustic elements that underlie voice quality perception. In this approach, listeners adjust synthesizer parameters to achieve a perceptual match to the original voice. This method of adjustment technique explicitly links the acoustic signal to voice quality. We hypothesize that this linkage across the speech chain increases the validity, reliability, and utility of the resulting perceptual responses, by providing listeners with an objective tool (a synthesizer) for quantifying what they hear. The proposed research focuses on four specific aims. First, we will expand, refine, and automate our synthesizer, to increase power, flexibility, and ease of use. Secondly, we will apply the synthesizer to examine the perceptual importance of several key aspects of pathologic voices, including the shape of the source spectrum, period doubling, and the perceptual interactions of harmonic and inharmonic (noise) energy in the voice source. In each case, we will determine which acoustic features are perceptually important and whether these features interact with others, and we will estimate just-noticeable-differences for the parameter as an index of the amount of acoustic change that is clinically meaningful on that dimension. Thirdly, we will evaluate clinicians' ability to apply the synthesizer as a tool for measuring voice quality in the clinic. We will also revisit the use of synthetic anchors in rating protocols as an alternative to the method of adjustment task. Finally, we will evaluate the external validity of the synthesis technique by comparing perceptual measures derived from sustained vowels to those derived from continuous speech. By increasing the power and ease of the synthesizer's use, examining listener sensitivity to important acoustic variables, and studying the generalizability of this approach, the proposed research will bring our goal of a standardized voice quality evaluation protocol within reach.