This is a documented descriptive, and cross-sectional study carried out at the Voice Laboratory of the Speech Therapy and Audiology Department of a university. It was evaluated and approved by the Research Ethics Committee of the institution, under Opinion n. 508200/2013.

This study used a set of synthesized voices developed by the VoiceSim synthesizer.33 The synthesizer consists of a computer system containing a vocal fold model and a representation of the vocal tract in the format of concatenated tubes, through which an acoustic wave propagates.32

Vocal deviations of roughness and breathiness were produced from the manipulation of acoustic parameters of fundamental frequency disturbance (flutter, tremor, and wow), additive noise and tension asymmetry between the vocal folds.33

Roughness was generated by manipulating the duration of the cycle of glottic excitation and jitter, with the introduction of a stochastic disturbance in the vocal fold tissue tension, using the formula: ΔK=αεK; where /α/ is a scale parameter, /¿/ is a random variable, and /K/ is a coefficient of vocal fold stiffness.

Breathiness was generated with the insertion of additive noise, according to the formula: Δμ=bεμ where /μ/ is the glottal airflow rate, /b/ is a scale parameter, and /¿/ is a random variable, similar to jitter.

The tension asymmetry parameters between the vocal folds, subglottic pressure and vocal fold separation were also controlled during the production of these synthesized signals. For more details on the synthesizer, please refer to the available literature.33

The speech material of the synthesized stimuli was the vowel /¿/ sustained for 3s. This vowel was chosen because it is commonly used in vocal and laryngeal evaluation procedures in Brazil,34 also considering that it is an oral, medium, open, and unrounded vowel, considered the most medium vowel of Brazilian Portuguese,34 which allows a more neutral and intermediate position of the vocal tract.

Therefore, 871 synthesized vocal signals were used, of which 426 (48.8%) were female and 446 (51.2%) were male signals, with different combinations of the previously mentioned acoustic parameters.

The acoustic analysis was performed using the VoxMetria software, version 4.5h, by CTS Informática (Pato Branco, Paraná, Brazil), in the vocal quality module. The PDD was used for this evaluation, in order to analyze the distribution of vocal signals according to area, quadrant, shape, and density.

Regarding the area, the software itself indicates whether the vocal signal is inside or outside the normal range (Fig. 1).

Figure 1.

Vocal signals inside (dots in blue color) and outside (dots in green color) the PDD normal area.

(0.16MB).

As for the quadrants, the PDD was divided into four equal quadrants17: lower left (1), lower right (2), upper right (3) and upper left (4) (Fig. 2).

Figure 2.

Division of PDD in quadrants.

(0.18MB).

Regarding the distribution of the points in relation to density (Figs. 3 and 4), the points concerning the distribution of the vocal signals were classified as concentrated, when the points were distributed inside a space corresponding to one square, or amplified, when the points were distributed throughout the space corresponding to more than one square of the PDD.

Figure 3.

Vocal sample with density concentrated on PDD.

(0.2MB).

Figure 4.

Vocal sample with density amplified in PDD.

(0.15MB).

The shape classification was performed using a simple 10-cm ruler on the printed sheet of each PDD generated by the software, corresponding to the image of each analyzed vocal signal, with no previous knowledge of the vocal deviation intensity and the predominant voice type.

The points concerning the distribution of vocal signals were categorized as vertical, when the distance between the points along the abscissa was lower than along the ordinate (X<Y); horizontal, when the distance between the points along the abscissa was higher along the ordinate (X>Y); and circular when the distance between the points along the ordinate and the abscissa was approximately the same (X≅Y).17

The perceptual-auditory evaluation session took place in a quiet environment and was performed by a speech therapist/audiologist who was also a voice specialist with more than 10 years of experience in this task.

The evaluator was instructed that voices should be considered normal when they were socially acceptable, naturally produced, without any irregularity, noise, or effort observable during the emission. The evaluator was also instructed that roughness would correspond to the presence of vibratory irregularity and breathiness would be associated with audible air escape during the emission. The evaluator was trained with anchor stimuli, containing normal emissions, and deviated ones at different degrees, as well as predominantly rough and breathy voices. Moreover, the evaluator was instructed about the cutoff values that would be used in this study,10 to categorize voices regarding the absence and presence of roughness and breathiness.

For the assessment, the evaluator used a Visual Analogue Scale (VAS), with a metric scale of 0–100mm, evaluating the intensity of vocal deviation (GD, general degree) and the roughness degree (RD) and breathiness degree (BD). The evaluation closest to 0 represents less vocal deviation, and the closer to 100, the greater the deviations.

For the assessment, each emission of the sustained vowel was presented three times through a speaker, at a comfortable intensity self-reported by the evaluator. At the end of the perceptual assessment session, 10% of the samples (88 signals) were randomly repeated for the evaluator's reliability analysis, using Cohen's Kappa Coefficient. The Kappa value was 0.88, indicating excellent reliability of the evaluator.35

In the current literature,10,36 distinct cutoff values are found for GD,36 RD10 and BD,10 used to categorize both the presence/absence of vocal deviation, and to classify the degree of the present deviation. Therefore, considering that the aim of this study is to investigate the performance of the PDD in the discrimination of the presence and degree of roughness and breathiness in synthesized voices, it was decided to use the cutoff values established for the classification of roughness and breathiness parameters.10

For RD, the following cutoff points are considered10: absence of roughness or Grade 0 (0–8.5mm), mild roughness or Grade 1 (8.6–28.5mm), moderate roughness or Grade 2 (28.6–59.5) and intense roughness or Grade 3 (≥59.6mm). In relation to BD, the following cutoff points were recommended: no breathiness or Grade 0 (0–8.5mm), mild breathiness or Grade 1 (8.6–33.5mm), moderate breathiness or Grade 2 (33.6–52.0mm) and intense breathiness or Grade 3 (≥ 52.1mm).

Thus, a correspondence was made between the VAS used for RD and BD and the numerical scale,10 as described below:

• Grade 0: RD and BD≤8.4mm;

• Grade 1: 8.5mm≤RD≤28.4mm and 8.5≤BD≤33.4mm;

• Grade 2: 28.5mm≤RD≤59.4mm and 33.5mm≤BD≤52.4mm;

• Grade 3: RD≥59.5mm and BD≥52.5mm.

The 8.4mm cutoff was also used to categorize the voices regarding the presence or absence of roughness and breathiness.10 Voices with values >8.4mm in RD and BD were considered as having the presence of roughness and breathiness in vocal emissions, respectively.

We chose not to analyze the tension parameter, since other studies have already shown that such characteristic is not specifically identified in the PDD,17,29 in addition to the lack of consensus regarding the inclusion of this parameter in the perceptual-auditory evaluation protocols.1,10

The GD evaluation36 was not used for signal categorization, but only for the sample characterization in the present study.

Therefore, based on the results of the perceptual-auditory analysis of the RD and BD, the following classification was observed:

• As for the presence of roughness: 128 (14.7%) signals without roughness (RD≤8.4mm) and 743 (85.3%) with roughness (RD≥8.5mm) (Table 1).

  Table 1.

  Distribution of vocal signals regarding the presence and degree of roughness and breathiness.

  Variable	n	%
  Degree of roughness
  Normal	128	14.70
  Mild to moderate	256	29.40
  Moderate	475	54.50
  Intense	12	1.40
  Total	871	100
  Degree of breathiness
  Normal	365	41.90
  Mild to moderate	187	21.50
  Moderate	310	35.60
  Intense	9	1.00
  Total	871	100

• As for the presence of breathiness: 365 (41.9%) signals without breathiness (BD≤8.4mm) and 506 (58.1%) with breathiness (BD≥8.5mm) (Table 1).

It is worth mentioning that a categorical analysis of the vocal quality predominant in the emission was not performed, but a same vocal signal could show roughness and breathiness components, since the criterion for the allocation of signals regarding the presence/absence of these components was the result of the independent evaluation of each of them through the VAS (RD and BD) and of the cutoffs established for these parameters (Table 2).

The statistical analysis was descriptive for all the assessed variables and Fisher's exact test and Chi-square test (x2) were used to compare the analysis of variables related to perceptual-auditory (presence and degree of roughness and breathiness) and acoustic measures (area, density, shape, and quadrant of the PDD). The Kruskal–Wallis test was used to compare the acoustic measurements according to the degree of roughness and breathiness. The level of significance was set at 5% for all analyses. The software used was the Statistical Package for Social Sciences (SPSS, version 21.0).

This study showed that the PDD area and quadrant were able to discriminate between normal signals and signals with roughness. Voices with roughness were predominantly located outside the area of normality and in the lower right quadrant.

Previous studies, carried out with adults’17 and children's voices,29 corroborate the findings obtained in the present study. Both the lower right quadrant and the PDD area were important to discriminate voices with presence and absence of roughness, showing these two parameters are robust and reliable to evaluate roughness in dysphonic and non-dysphonic voices.

The PDD evaluates signal irregularity in its horizontal position, being associated to the concept of roughness.24,26 The greater the irregularity of the vocal signal, the greater its displacement from left to right in the chart. This fact justifies the location of rough voices outside the area of normality and in the lower right quadrant, both in the present study and in previous ones.17,29

Additionally, it is emphasized that roughness is one of the universal parameters of the perceptual-auditory evaluation of vocal quality, representing an important characteristic in the identification of the presence of vocal or laryngeal alterations.37

Roughness is commonly related to the presence of structural and/or functional alterations in the larynx, such as is seen in cases of edema, vascular dysgenesis, nodular lesions, polyps, or any other component that generates a mass increase in the membranous portion of the vocal folds38 and, consequently, irregularities in the vocal fold vibratory pattern. In the acoustic plane, roughness is associated to the jitter and shimmer parameters.19

As for the distribution of voices with different degrees of roughness in the PDD, it was verified that vocal signals with a greater roughness component were proportionally outside the area of normality and in the lower right quadrant. Regarding density, signals with moderate and intense deviation predominantly showed concentrated density.

It is noteworthy that 35.93% (n=46) of the synthesized voices without roughness were outside the area of normality, whereas 12.10% (n=31) of the voices with mild-to-moderate degree of roughness were inside the area of normality, that is, the PDD showed a greater confounding factor in the identification of voices without roughness, with a slight deviation in relation to the signals with a higher degree of roughness (moderate and intense).

In traditional models, with the use of algorithms that extract isolated jitter and shimmer measurements, an inverse behavior is observed, as the use of these isolated measures is less reliable in the evaluation of more deviant voices.15,17,20,24,26,39–41

Regarding density, few studies17,28,29 specifically included this parameter for PDD analysis and none of them investigated the distribution of voices with different degrees of roughness as a function of PDD density. Only one of these studies17 showed a difference in the distribution of signals with and without vocal deviation regarding density, with the deviated signals characterized as having amplified density.

In other studies where PDD was used,20,24,26,40–42 the density parameter can be inferred from the distance between the points only on the abscissa axis, being associated with signals with amplified or concentrated density, respectively. All these studies were longitudinal ones and produced a tendency for less dispersion of the points on the post-intervention abscissa axis, although there is great individual variability in this parameter throughout the treatment,26 with significant differences being observed only between pre- and post-treatment conditions.

This study showed greater variability in the distribution of the signals without a roughness component or with a mild-to-moderate degree of roughness between the concentrated and amplified densities. This fact confirms the good performance of the PDD in analyzing signals with a wide range of deviation and its reliability in the assessment of the most deviant signals. Additionally, it can be inferred that the PDD density parameter seems to be more robust to qualitatively analyze the patient's evolution regarding the roughness component in vocal emission.

Regarding the shape, although a statistical significance was verified, a distribution pattern of the signals with different degrees of roughness as a function of this PDD parameter was not observed. In all grades, the voices were predominantly horizontal, with differences being observed only between the proportions of the groups. This finding corroborates the literature, as there is a tendency for the signals to show a predominance of the dispersion of the points in the horizontal dimension, regardless of the presence and degree of vocal deviation.20,24,26,40–42

Even in the original proposal for the classification of vocal signals as a function of the PDD shape, no significant difference was observed between healthy and deviant signals, as well as between different degrees of deviation and between rough, breathy, and tense voices.17 Therefore, the shape of the points distributed in the PDD does not seem to be a robust parameter for signal differentiation.

When comparing the distribution of vocal signals with and without breathiness as a function of the PDD parameters, it was observed that area and quadrant were able to discriminate normal vocal signals from breathy ones. Breathy vocal signals were outside the normal range and were predominantly located in the lower right quadrant.

Breathiness is among the universally accepted parameters for the perceptual-auditory evaluation of vocal quality and for the characterization of a dysphonic voice.4,8,37 Thus, the fact that the PDD correctly identifies the breathy signals outside the area of normality reinforces its usefulness in the clinical context of vocal assessment.

However, it was observed that the PDD area and quadrant parameters showed identical behavior, in both rough and breathy voices. The vocal signals with roughness and breathiness were found outside the area of normality and in the lower right quadrant. Therefore, one can discuss the interrelationships of these two parameters in physiological and perceptual terms.

The presence of breathiness is physiologically associated with a higher degree of separation between the vocal processes, lower convexity of the free edge of the vocal folds and the shorter time of the closed phase of the glottic cycles43 In turn, vocal folds that are further away from the midline tend to vibrate with greater irregularity and less amplitude of the mucosal wave,44 which, consequently, generates the roughness component in the emission.37

Therefore, considering that the signals with roughness and breathiness showed, in general, moderate deviation, with GD of 62.19±14.80 and 65.28±14.75 points in the VAS,36 respectively, one understands the similar distribution of signals with roughness and breathiness in the PDD area and quadrant.

Although the synthesizer used to generate the signals in this study allows the creation of voices with isolated components of roughness (disturbance) and breathiness (additive noise), this separation was not used in the present study. We suggest further investigations with separation of the exclusively rough and breathy signals to assess the performance of the PDD in this classification.

In other studies,17,29 the breathy voices were located outside the area of normality, but were distributed between the lower right and upper right quadrants. Some methodological issues need to be highlighted to evidence the similar distribution of the rough and breathy voices in the lower right quadrant in this study.

The two aforementioned studies17,29 used as a criterion to classify the voices as rough, breathy, or tense, a forced choice task, in which the evaluator, if he/she considered the emission deviant, should determine the predominant vocal quality. This type of evaluation task allows only one possibility of choice for each emission and not necessarily a classification regarding the presence/absence of each deviated parameter in the emission.

In turn, the present study evaluated the degree of roughness and breathiness present in the emission through a VAS. Based on the cut-off values, the presence/absence of such components was established, with the possibility that the same signal would concomitantly show the presence of one or more of them, which is close to the usual conditions of deviant vocal production.

Another finding of this research is the high percentage of voices without breathiness (75.40%; n=276) classified outside the normal range of the PDD. In a qualitative data analysis, it can be observed that the GD of deviation of these signals is 53.35±16.49. Therefore, although these signals did not show auditory-perceived breathiness, they were probably evaluated as deviated in the VAS due to the presence of roughness in the emission.

When comparing the results regarding the proportion of voices with presence/absence of roughness and presence/absence of breathiness identified inside and outside the PDD normality area, it is observed that there is a greater identification of voices without roughness within the area of normality (64.07%, n=82) and a greater identification of voices without the breathiness component outside the normality area (75.35%; n=275).

Qualitatively, a difference of more than 20 points was found regarding the VAS GD between voices without roughness and without breathiness, with higher GD values in the latter group. This difference in itself would justify the results regarding the higher proportion of signals without the breathiness component identified outside the normal range.

These findings reinforce that, even in conditions where the perceptual-auditory evaluation criteria used to classify the signals were not intended to maximize the differences between them, but to evaluate them over a continuum, the PDD was also efficient for vocal evaluation, mainly in relation to the most deviant signals.

It is suggested that other studies be carried out using the same methodology and criteria of perceptual-auditory evaluation used in this study, adding to them the criterion that the signals selected for investigation have only one of the components deviated from the cutoff values of the VAS.

Regarding the degree of breathiness, there was a difference in the distribution of the signals as a function of the PDD area, density, and quadrants. It was observed that the higher the degree of breathiness, the greater the proportion of signals located outside the area of normality, in the lower right quadrant and with concentrated density. Therefore, it is verified that the greater the breathiness component in the vocal signal, the greater the capacity of the PDD to correctly identify the presence of the deviation.

As previously mentioned, such finding regarding the classification of signals with higher degree of deviation constitutes one of the greatest advantages of the PDD, as it fills an existing gap15 regarding the use and reliability of traditional measures of disturbance and noise in the evaluation of voices with moderate and intense deviations.

Once again, a similar distribution of the voices with different degrees of roughness and breathiness was observed as a function of the area, quadrant, and density of the PDD. The only difference between the voices with different degrees of roughness and breathiness is the distribution of the signals with Grade 2, in which there was a higher level of correct identification of the group of voices without roughness within the PDD normality area. This fact has already been discussed in this section.

The vertical axis of the PDD evaluates the presence of additive noise in the vocal signal, compatible with the presence of the breathiness component.26 Therefore, it was expected that the higher the breathiness component in the emission, the greater the proportion of signals toward the upper left quadrant.

In the study17 with voices of dysphonic adults, it was observed that breathy voices, although they were predominantly distributed in the upper left quadrant (52.6%; n=30); 19.3% (n=11) were also situated in the lower right quadrant. With the pediatric population,29 breathy voices were distributed in the lower right (35%, n=7), lower left (30%, n=6), upper right (30%, n=6) and upper left (5%, n=1) quadrants.

In studies26,41 with patients presenting with unilateral vocal fold paralysis26 and individuals with bilateral vocal fold paralysis,26,41 it was found that only the second group, whose patients showed intense breathiness, had their voices located in the upper right quadrant. In turn, individuals with unilateral paralysis had their voices distributed between the lower left and lower right quadrants.26

In general, in high lesions of the vagus nerve, the vocal folds are more distant from the midline and the vocal emission does not originate from the glottic vibration mechanism, but comes primarily from the turbulent transglottic airflow and its propagation in the vocal tract,45,46 which would justify the presence of these signals in the upper right quadrant.26

In the present study, only nine signals were classified as having severe breathiness deviation, and of these, only one (11.11%) was in the upper right quadrant. In this way, two points can be highlighted: first, the sample size, since a different result could have been observed in this distribution with a larger sample of breathy voices with intense deviations; second, as already emphasized in the discussion, there is an overlap of the type of vocal deviation in the assessed signals, since the presence of only one type of deviation in each emission was not used as eligibility criterion.