Much previous research has demonstrated that listeners do not agree well when using traditional rating scales to measure pathological voice quality. Although these findings may indicate that listeners are inherently unable to agree in their perception of such complex auditory stimuli, another explanation implicates the particular measurement method-rating scale judgments-as the culprit. An alternative method of assessing quality-listener-mediated analysis-synthesis-was devised to assess this possibility. In this new approach, listeners explicitly compare synthetic and natural voice samples, and adjust speech synthesizer parameters to create auditory matches to voice stimuli. This method is designed to replace unstable internal standards for qualities like breathiness and roughness with externally presented stimuli, to overcome major hypothetical sources of disagreement in rating scale judgments. In a preliminary test of the reliability of this method, listeners were asked to adjust the signal-to-noise ratio for 12 synthetic pathological voices so that the resulting stimuli matched the natural target voices as well as possible For comparison to the synthesis judgments, listeners also judged the noisiness of the natural stimuli in a separate task using a traditional visual-analog rating scale. For 9 of the 12 voices, agreement among listeners was significantly (and substantially) greater for the synthesis task than for the rating scale task. Response variances for the two tasks did not differ for the remaining three voices. However, a second experiment showed that the synthesis settings that listeners selected for these three voices were within a difference limen, and therefore observed differences were perceptually insignificant. These results indicate that listeners can in fact agree in their perceptual assessments of voice quality, and that analysis-synthesis can measure perception reliably.

Although perceptual assessment is included in most protocols for evaluating pathologic voices, a standard set of valid scales for measuring voice quality has never been established. Standardization is important for theory and for clinical acceptance, and also because validation of objective measures of voice depends on valid perceptual measures. The present study used large sets (n = 80) of male and female voices, representing a broad range of diagnoses and vocal severities. Eight experts judged the dissimilarity of each pair of voices, and responses were analyzed using nonmetric individual differences multidimensional scaling. Results indicate that differences between listeners in perceptual strategy are so great that the fundamental assumption of a common perceptual space must be questioned. Because standardization depends on the assumption that listeners are similar, it is concluded that efforts to standardize perceptual labels for voice quality are unlikely to succeed. However, analysis by synthesis may provide an alternate means of modeling quality as a function of both voices and listeners, thus avoiding this problem.

From a sample of young male Californians, ten speakers were selected whose voices were approximately normally distributed with respect to the "easy-to-remember" versus "hard-to-remember" judgments of a group of raters. A separate group of listeners each heard one of the voices, and, after delays of 1, 2, or 4 weeks, tried to identify the voice they had heard, using an open-set, independent-judgment task. Distributions of the results did not differ from the distributions expected under the hypothesis of independent judgments. For both "heard previously" and "not heard previously" responses, there was a trend toward increasing accuracy as a function of increasing listener certainty. Overall, heard previously responses were less accurate than not heard previously responses. For heard previously responses, there was a trend toward decreasing accuracy as a function of delay between hearing a voice and trying to identify it. Information-theoretic analysis showed loss of information as a function of delay and provided means to quantify the effects of patterns of voice confusability. Signal-detection analysis revealed the similarity of results from diverse experimental paradigms. A "prototype" model is advanced to explain the fact that certain voices are preferentially selected as having been heard previously. The model also unites several previously unconnected findings in the literature on voice recognition and makes testable predictions.

Although jitter, shimmer, and noise acoustically characterize all voice signals, their perceptual importance in naturally produced pathological voices has not been established psychoacoustically. To determine the role of these attributes in the perception of vocal quality, listeners were asked to adjust levels of jitter, shimmer, and the noise-to-signal ratio in a speech synthesizer, so that synthetic voices matched naturally produced tokens. Results showed that, although listeners agreed well in their judgments of the noise-to-signal ratio, they did not agree with one another in their chosen settings for jitter and shimmer. Noise-dependent differences in listeners' ability to detect changes in amounts of jitter and shimmer implicate both listener insensitivity and inability to isolate jitter and shimmer as separate dimensions in the overall pattern of aperiodicity in a voice as causes of this poor agreement. These results suggest that jitter and shimmer are not useful as independent indices of perceived vocal quality, apart from their acoustic contributions to the overall pattern of spectrally shaped noise in a voice. (c) 2005 Acoustical Society of America.

Traditional interval or ordinal rating scale protocols appear to be poorly suited to measuring vocal quality. To investigate why this might be so, listeners were asked to classify pathological voices as having or not having different voice qualities. It was reasoned that this simple task would allow listeners to focus on the kind of quality a voice had, rather than how much of a quality it possessed, and thus might provide evidence for the validity of traditional vocal qualities. In experiment 1, listeners judged whether natural pathological voice samples were or were not primarily breathy and rough. Listener agreement in both tasks was above chance, but listeners agreed poorly that individual voices belonged in particular perceptual classes. To determine whether these results reflect listeners' difficulty agreeing about single perceptual attributes of complex stimuli, listeners in experiment 2 classified natural pathological voices and synthetic stimuli (varying in f0 only) as low pitched or not low pitched. If disagreements derive from difficulties dividing an auditory continuum consistently, then patterns of agreement should be similar for both kinds of stimuli. In fact, listener agreement was significantly better for the synthetic stimuli than for the natural voices. Difficulty isolating single perceptual dimensions of complex stimuli thus appears to be one reason why traditional unidimensional rating protocols are unsuited to measuring pathologic voice quality. Listeners did agree that a few aphonic voices were breathy, and that a few voices with prominent vocal fry and/or interharmonics were rough. These few cases of agreement may have occurred because the acoustic characteristics of the voices in question corresponded to the limiting case of the quality being judged. Values of f0 that generated listener agreement in experiment 2 were more extreme for natural than for synthetic stimuli, consistent with this interpretation.

The validity of perceptual measures of vocal quality has been neglected in studies of voice, which focus more commonly on rater reliability. Validity depends in part on reliability, because an unreliable test does not measure what it is intended to measure. However, traditional measures of rating reliability only partially represent interrater agreement, because they cannot reflect variations or patterns of agreement for specific voice samples. In this paper the likelihood that two raters would agree in their ratings of a single voice is examined, for each voice in five previously gathered data sets. Results do not support the continued assumption that traditional rating procedures produce useful indices of listeners' perceptions. Listeners agreed very poorly in the midrange of scales for breathiness and roughness, and mean ratings in the midrange of such scales did not represent the extent to which a voice possesses a quality, but served only to indicate that listeners disagreed. Techniques like analysis by synthesis or judgment of similarity avoid decomposing quality into constituent dimensions, and do not require a listener to compare an external stimulus to an unstable internal representation, thus decreasing the error in measures of quality. Modeling individual differences in perception can increase the variance accounted for in models of quality, further reducing the error in perceptual measures. Thus such techniques may provide valid alternatives to current approaches.

Although the terms "breathy" and "rough" are frequently applied to pathological voices, widely accepted definitions are not available and the relationship between these qualities is not understood. To investigate these matters, expert listeners judged the dissimilarity of pathological voices with respect to breathiness and roughness. A second group of listeners rated the voices on unidimensional scales for the same qualities. Multidimensional scaling analyses suggested that breathiness and roughness are related, multidimensional constructs. Unidimensional ratings of both breathiness and roughness were necessary to describe patterns of similarity with respect to either quality. Listeners differed in the relative importance given to different aspects of voice quality, particularly when judging roughness. The presence of roughness in a voice did not appear to influence raters' judgments of breathiness; however, judgments of roughness were heavily influenced by the degree of breathiness, the particular nature of the influence varying from listener to listener. Differences in how listeners focus their attention on the different aspects of multidimensional perceptual qualities apparently are a significant source of interrater unreliability (noise) in voice quality ratings.