If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Discipline of Speech Pathology, Faculty of Medicine and Health, Sydney School of Health Sciences, The University of Sydney, Sydney, AustraliaDoctor Liang Voice Program, Faculty of Medicine and Health, Sydney School of Health Sciences, The University of Sydney, Sydney, Australia
Discipline of Speech Pathology, Faculty of Medicine and Health, Sydney School of Health Sciences, The University of Sydney, Sydney, AustraliaDoctor Liang Voice Program, Faculty of Medicine and Health, Sydney School of Health Sciences, The University of Sydney, Sydney, Australia
Discipline of Speech Pathology, Faculty of Medicine and Health, Sydney School of Health Sciences, The University of Sydney, Sydney, AustraliaDoctor Liang Voice Program, Faculty of Medicine and Health, Sydney School of Health Sciences, The University of Sydney, Sydney, Australia
Address correspondence and reprint requests to Catherine Madill, Faculty of Medicine and Health, Sydney School of Health Sciences, The University of Sydney, Sydney, Australia.
Discipline of Speech Pathology, Faculty of Medicine and Health, Sydney School of Health Sciences, The University of Sydney, Sydney, AustraliaDoctor Liang Voice Program, Faculty of Medicine and Health, Sydney School of Health Sciences, The University of Sydney, Sydney, Australia
Evidence across a range of musically trained, hearing disordered and voice disordered populations present conflicting results regarding the relationship between pitch discrimination (PD) and voice quality. PD characteristics of female speakers with and without a musical training background and no self-reported voice disorder, and the relationship between PD and voice quality in this particular population, have not been investigated.
Aims
To evaluate PD characteristics in a cohort of female participants without a self-reported voice disorder and the relationship between PD and acoustic voice measures.
Method
One hundred fourteen female participants were studied, all of whom self-reported as being non-voice disordered. All completed the Newcastle Assessment of Pitch Discrimination which involved a two-tone PD task. Their voices were recorded producing standardized vocal tasks. Voice samples were acoustically analyzed for frequency-domain measures (fundamental frequency and its standard deviation, and harmonics-to-noise ratio) and spectral-domain measures (cepstral peak prominence and the Cepstral/Spectral Index of Dysphonia). Data were analyzed for the whole cohort and for musical and non-musical training backgrounds.
Results
In the whole cohort, there were no significant correlations between PD and acoustic voice measures. PD accuracy in musically trained speakers was better than in non-trained speakers and correlated with fundamental frequency standard deviation in prolonged vowel tasks. Vocalists demonstrated superior PD accuracy and fundamental frequency standard deviation in prolonged vowels compared to instrumentalists but did not show significant correlations between PD and acoustic measures. The Newcastle Assessment of Pitch Discrimination was a reliable tool, showing moderate-good prediction value in differentiating musical background.
Conclusions
There was little evidence of a relationship between PD and acoustic measures of voice quality, regardless of musical training background and superior PD accuracy among the musically trained. These data do not support ideas concerning the co-development of perception and action among individuals identified as having voice quality measures within normal ranges. Numerous measures of voice quality, including measures sensitive to pitch, did not distinguish across musically and non-musically trained individuals, despite individual differences in pitch discrimination.
This dependency of voice production upon auditory perception has been shown in the use of auditory feedback for learning and self-monitoring in singing and vocal training
Relationships between perception and production can be the result of bidirectional influences, including the effects of perception on production and the effects of production on perception. Pitch discrimination (PD), that is, the ability to tell the difference between pitches, has been frequently used to test auditory perception function.
The role of auditory perception in voice production
The influence of perception on voice/speech production can be demonstrated typically in the hearing impaired. Habitual vocal production of individuals with hearing impairments is often associated with reduced pitch control, such as a lack of pitch variation or excessive abnormal pitch variations.
Moderate-profound hearing loss has also been associated with abnormal perceptual voice quality such as breathiness, hoarseness, and higher speaking fundamental frequency (F0).
Whilst this unidirectional link between auditory abilities and vocal production abilities has been observed, it is unknown whether a deficit in central auditory processing or a degraded peripheral auditory signal were the key factors causing reduced voice quality and control.
A relationship is also demonstrated in evidence that musicians have demonstrated better pitch production accuracy than non-musicians.
Smaller difference limens for frequency (DLFs), indicative of better PD, were correlated with better pitch matching vocal accuracy in highly trained, working musicians.
However, correlations were not observed in non-musicians. In addition, in inaccurate singers, no relationship has been observed between PD and pitch production ability.
showed that patients with hyperfunctional voice disorders had poorer PD compared to speakers without a voice disorder. The changes in PD have been considered as possibly playing a role in pathogenesis of behavioral voice disorders. Stepp et al
found that patients with hyperfunctional voice disorders showed different adaptive responses in pitch perturbation tasks compared to controls. Muscle Tension Dysphonia patients were found to have a significantly larger magnitude adaptive response to changes to auditory feedback as compared to people without dysphonia, suggesting a possible link between pitch perception and voice production.
Pitch pattern recognition abilities as assessed in three-tone sequences were poorer in females with voice disorders in comparison to females without disorders and was also strongly correlated to reduced vocal reproduction of musical tones.
The effects of voice production on auditory perception
There is limited understanding of the effects of voice production on auditory perception. Previous research has shown auditory-perception difficulties in patients with vocal dysfunction, suggesting that impairments in voice production can negatively impact on perception. For example, patients with unilateral vocal fold paralysis (UVFP) showed differences in neural areas associated with vocal-motor function,
however, the direction of causality of auditory perceptual abilities and voice disorder is not evident in this study. The “Linked Dual Representation Model” suggests that vocal production can be mediated by high-level awareness and perception, and low-level (non-conscious) perception.
On the one hand, pitch perception abilities would be linked to production through training, such that training in pitch perception transfers to voice production. On the other hand, pitch perception and voice production are thought to be linked through non-aware production pathways. In the latter case, adaptations in voice are made online due to low-level auditory feedback-based corrections.
Some other authors, however, have not observed a significant correlation between PD and voice. Davis and Boone
did not find significant differences between the two groups when comparing PD and tonal memory between patients with hyperfunctional voice disorders and control speakers. Murray et al
found no relationship between PD and voice production in children with and without vocal nodules.
Measurement of auditory perception
Pitch discrimination (PD) has been assessed in isolation using pitch-based tasks, such as two-tone PD tasks in varying frequency discrimination protocols when comparing trained to untrained musicians and vocalists. Another popular laboratory-based method of measuring pitch perception is known as identifying the just noticeable difference
Same or different pitch? Effects of musical expertise, pitch difference, and auditory task on the pitch discrimination ability of musicians and non-musicians.
An adaptive tracking procedure is used to obtain a DLF, whereby pitch differences are reduced following a specified number of correct responses and increased followed by one incorrect response.
Whilst these protocols are successful in identifying the smallest pitch difference that individuals can recognize, the process is long, complex, and the equipment required to conduct these tests is not readily available in clinical settings. As such, two-tone PD tasks may provide the most direct and simple measure of the auditory perceptual system.
Voice quality has a complex and oft described relationship with signal frequency characteristics and thus can affect perception of pitch. Voice quality descriptors such as roughness, breathiness and hoarseness are associated with acoustic measures such as jitter, which quantify the amount of variations in the fundamental frequency (F0) in prolonged vowel tasks.
Given that small variations in fundamental frequency can affect perceptions of voice quality, some acoustic measures of voice quality may be more or less appropriate and possibly more sensitive to fine voice-motor control than others. Similarly, voice quality may also influence perception of pitch. Collecting both voice quality data and auditory perception of PD data in the same population can help to clarify whether such relationships exist and what this might mean, based on musical background, for transfer pathways that are more conscious and explicit. Therefore, we included a range of acoustic voice production measures in our study that relate to not only pitch, but perception of voice quality.
To date, in no study has the relationship between acoustic voice measures and PD been investigated in a large non-clinically disordered cohort where differences between individuals in both voice and PD
are apparent but not clinically significant. Specifically, it is unclear whether frequency-domain measures (fundamental frequency and its standard deviation) and spectral-domain measures related to voice quality are correlated with PD accuracy. The aims of the present study were to: 1) Investigate PD characteristics in a cohort of female speakers who were self-identified as non-voice disordered; and 2) Investigate acoustic voice characteristics and their relationship to PD in a cohort of female speakers self-identified as having no voice problems. Overall, we hoped to better evaluate the inherent and/or trained links between pitch perception and various voice production measures to help understand the co-development or otherwise of these skills.
MATERIAL AND METHODS
Ethical approval
The study protocol was approved by The University of Sydney Human Research Ethics Committee (Protocol number: 2016/1001). Written informed consent was obtained from all participants to partake in this study. The study was implemented in accordance with relevant ethical guidelines and regulations.
Participants
There were 114 female participants with a mean age of 23.1 years, standard deviation, SD = 3.8, range = 18-40 years. All were first or second language English-speaking university students. Inclusion criteria included: 1) No self-reported or previous diagnosis of voice disorder; 2) Normal hearing (passing 20-decibel threshold in a pure-tone audiometric screening on frequencies of 1kHz, 2kHz and 4kHz); 3) Non-smokers; 4) Did not regularly use inhaled corticosteroids (a common medication for asthma known to impact voice quality); and 5) Had not experienced upper respiratory problems within 2 weeks before the study.
Participants completed a case history questionnaire to determine history of voice disorders, smoking, upper respiratory problems, language backgrounds, musical background, and voice/musical training. They also completed the Screening Index for Voice Disorders
Table 1 shows the voice disorder screening results using the SIVD and the VHI-10 scores. All participants had a SIVD score of 3 or below (where 5 or above indicates voice disorder
). Although the majority of people had a VHI-10 score below 7.5, there were 15 who had a score above (which is the cut-off value for determination of voice quality handicap).
Note that cut-off value refers to the clinical reference value for determination of a mild voice disorder (along with a publication reference, Ref) and scale maximum (max) values.
Abbreviations: HNR, harmonics-to-noise ratio in dB; NA, not available; SD, standard deviation; SIVD, Screening Index for Voice Disorders; VHI-10, Voice Handicap Index – 10.
Participants were classified based on musical background as determined from the case history questionnaire. Musically trained individuals were defined as those having formally learnt a musical instrument for at least a year past the age of 5 years old in individual lessons or in band rehearsals. This resulted in 58 participants defined as musically trained and 56 as non-musically trained. A small proportion of the musicians were identified as having voice training at some point during their developmental years or during adulthood; either individual lessons or in group choir (n = 10). Due to the potential significance of differences between these individuals and instrumentalists (where one group has training in voice and the other does not), secondary comparative analyses were run on these groups.
Pitch discrimination tasks
We used the Newcastle Assessment of Pitch Discrimination (NeAP),
which is a two-tone discrimination task. Auditory stimuli were computer-generated tones played on a Dell computer (Optilex 760) via a speaker system (Harman/Kardon HK645) calibrated to 65-65.2 dBA hearing level. Hearing level was measured at ear level using lingWAVES SPL meter II model IEC 651. Participants completed the default protocol that was pre set on the NeAP software. They were instructed to listen to two tones and to indicate which tone was higher in pitch or if they were the same. Responses were provided by clicking on the relevant icon on the computer screen (‘1’, ‘2’ or ‘same’). The default protocol contained twenty pairs of sine waves presented twice for a total of forty tone pairs; tones were presented in random order. Frequencies ranged between B2 (123.47 Hz) to D#4 (311.13 Hz) and pitch differences between tones ranged between one tone and a third of a semitone (Appendix 1). There was an average completion time of ∼5 minutes. Participants completed the PD task before completing their voice recordings. The percentages of accurate responses were calculated for each tone pair (t = 20) by dividing the number of accurate responses by the total responses for that tone pair. All tone pairs (total 20) were presented a second time in random order in the same session for reliability analyses.
Voice recordings
Voice recording took place in a soundproof booth. Participants were fitted with a calibrated, head-mounted, cardiod condenser microphone AKG C520 placed 5 cm and 45° away from the centre of their mouth. The microphone was calibrated using a single sine wave stimulus with frequency of 333.3 Hz at an average intensity of 60 dBA. Voice recordings were made using a Layla 24/96 Multitrack Recording System and Adobe Audition software (Version 1.5)
and to sustain the vowel ‘ah’ for 6 seconds. Participants were recommended to use their most comfortable pitch and loudness in producing these vocal tasks. The Rainbow Passage was selected for acoustic analysis as it is a phonemically balanced paragraph.
The use of prolonged vowel tasks ensures measurement of vocal perturbation and glottal noise are more accurate than in continuous speech tasks, where these measures are often influenced by intonation and other effects.
All vocal tasks (sustained vowel, CAPE-V phrases, and Rainbow Passage) were repeated for a total of three trials and an average of the three trials was used for statistical analysis.
Acoustic voice analysis
Prolonged vowel samples of the vowel ‘ah’ were trimmed to include two seconds of phonation in the middle section of the voice signal. Samples excluded the first and last one-second region of the signal, as these regions have been reported to contain the highest signal perturbation.
Only type 1 and 2 signals were included in the final analysis for frequency-based analyses (such as harmonics-to-noise ratio, HNR) due to their enhanced reliability compared to type 3 and 4 signals.
voice recordings were edited to include only the second and third sentences “…The rainbow is a division of white light into many beautiful colours. These take the shape of a long round arch, with its path high above, and its two ends apparently beyond the horizon…”. This task was used to allow comparisons with other studies on cepstral/spectral measures.
Connected speech samples were trimmed to have less than 0.5 seconds’ silence at the beginning and end of each signal in preparation for acoustic analysis.
Frequency-based acoustic measurements
Frequency-based analysis was performed to obtain acoustic values for F0, standard deviation of F0 (F0SD), HNR, and intensity. These were measured using default settings of the acoustic analysis program Praat version 6.0.25.
In addition, the F0 used in a prolonged vowel is arbitrary as participants were instructed to sustain the vowel ‘ah’ at a comfortable pitch for 6 seconds without any verbal model or reference.
Spectral-based measurements
Two spectral-based measures were included in this study: Smoothed Cepstral Peak Prominence (CPPS) and Cepstral/Spectral Index of Dysphonia (CSID). CPPS is obtained from a log power spectrum of a log power spectrum in which its quefrency at the cepstral peak represents the fundamental period of the signal.
This measure was included as it is a measure of regularity of the fundamental period with periodic signals showing more prominent cepstral peak (well-defined F0) than aperiodic signals.
for vowel (CSIDv) and manually calculated for the Rainbow Passage (CSIDrp) based on CPP, Low/High Spectral Ratio (LH), and standard deviation (SD) of Low/High Spectral Ratio measured in ADSV using the following formula
The CSID is a multifactorial measure incorporating the means and standard deviations of Cepstral Peak Prominence and ratio of low versus high frequency spectral energy to provide a quantitative measure for dysphonia.
An extended four second prolonged vowel signal was used to calculate CSID due to limitations in the use of the initial two second signal.
Auditory perceptual voice analyses
The auditory perceptual analyses allowed us to rate voice quality of the whole cohort and identify those with potential voice abnormality. The 3 s sustained /a/ vowel and the Rainbow Passage of the second trial were extracted and combined into a separate audio file for each participant. These files were randomized and uploaded onto Bridge2Practice.com
for auditory-perceptual assessment. Voice samples were presented in Bridge2Practice.com in random order across raters. Four voice professionals (three speech-language pathologists and one ear, nose and throat specialist doctor) listened to audio samples as many times as they wished and judged the severity of dysphonia by placing a slider on a 100-point rating scale in an online rating form as per the Consensus Auditory-Perceptual Evaluation of Voice (CAPE-V) rating form.
An average of the four responses were used for statistical analyses.
Thirty percent of voice samples (35 files) were duplicated, randomized throughout, and rated during the same session to evaluate intra- and inter-rater reliability using intraclass correlation coefficients (ICC). For inter-rater reliability, average ICC was 0.68 (95% CI = 0.56-0.77, P < 0.01). Intra-rater reliability for each of the raters ranged from 0.69 to 0.83 (95% CI = 0.36-0.92, P < .003). This was considered acceptable for the study as these ICC values would be considered moderate-to-good correlations.
Data were checked for normal distribution before selecting comparison and correlation tests. When data were normally distributed, subgroup comparisons were made using independent samples t tests (Cohen's d is also provided as a measure of effect size). A P value < 0.05 was used to determine statistical significance in group comparison analyses. Pearson's correlation coefficient was used to assess the relationship between individual acoustic voice measures and PD accuracy. Where there were multiple correlation calculations, Bonferroni's adjustment was implemented to minimize Type I error. Bonferroni-adjusted p was deemed significant if it was < 0.05. Receiver Operating Characteristic (ROC) curve analysis was used to calculate the sensitivity and specificity of the NeAP in differentiating musical from non-musical training groups. The significance level was P < 0.05 as P values that are too small can lead to Type II errors.
RESULTS
Reliability, sensitivity, and specificity of the NeAP
As the NeAP has not been used clinically, it was important to test the validity of this tool. The intra-class correlation coefficient (ICC) was calculated to assess reliability of agreement across tone pairs, based on the entire cohort. Correlations were all high and positive, both between single measures (ICC = 0.853, 95% CI = 0.794-0.896, P < 0.001) and average measures (ICC = 0.921, 95% CI = 0.885-0.945, P < 0.001).
ROC curve was calculated from NeAP scores with musical background being the defining variable of positive (n = 58) and negative (n = 56). The NeAP could differentiate the two groups with reasonable balance (highest sensitivity with highest specificity) at a sensitivity of 74.14% and specificity of 73.21%. The optimal cut-off point was identified with Youden J index
of 0.474 at PD score ≥ 75%. The area under the ROC curve (AUC) is shown in Figure 1 [95% CI = 0.690-0.850, Z = 6.372, P < 0.001] Figure 1. shows that this NeAP (default protocol) had moderate-good prediction accuracy in differentiating musical training background.
FIGURE 1ROC curve of percentage of correct response of NeAP. AUC, area under the ROC curve.
PD accuracy was calculated as the percentage of correct responses from the 20 tone pairs. Although normative data were not currently available for the NeAP test, PD was generally high (above 70%).
Pitch discrimination accuracy in musically trained and non-trained groups
Mean (SD) of PD accuracy (%) was 71.45 (21.11), 81.29 (16.32), and 61.25 (20.76) for the whole sample, musically trained group, and non-musically trained groups, respectively. The musically-trained group had a significantly higher percentage of accurate responses than the non-trained group, t(104.35) = 5.72, P < 0.001, d = 1.12. The maximal pitch difference between tone pairs that were incorrect showed that the non-trained group made errors with larger pitch differences than the musically trained group, t(100) = 3.02, P = 0.003, d = 0.60; as shown in Figure 2.
In the non-trained subgroup, vocal intensity in the Rainbow Passage was correlated with both Rainbow Passage CPPS, r = 0.61 and Vowel HNR, r = 0.46 (both Ps < 0.001). No correlations between vocal intensity, CPPS and HNR were observed in the mildly voice disordered and non-disordered subgroups.
FIGURE 2Maximal pitch difference of incorrect tone pairs for the musically trained and non-trained group (note error bars show 95% confidence intervals).
Pitch discrimination accuracy in instrumentalist and vocalist groups
Mean (SD) of PD accuracy (%) for instrumentalist and vocalist group was 79.90 (17.3) and 88.00 (7.89), respectively. The vocalist group had a significantly higher percentage of accurate responses than the instrumentalist group, t(30.27) = 2.23, P = 0.03, d = 0.83. Analysis of the maximal pitch difference between tone pairs that were incorrect showed that the instrumentalist group made errors with larger pitch differences (mean = 12.2, SD = 9.9 Hz) than the vocalist group (mean = 5.3, SD = 1.4 Hz), t(45.527) = 4.305, P < 0.001, d = 1.585.
Acoustic voice characteristics
Acoustic voice measures for the whole cohort, along with normative comparative data for the main measures are displayed in Table 2. In general, the sample showed acoustic voice characteristics that were within published normal ranges. Mean CPPS values were higher than the cut-off values for individuals with dysphonia.
Comparison of cepstral peak prominence measures using the ADSV, SpeechTool, and VoiceSauce acoustic analysis programs in vocally healthy female speakers.
Mean CSID scores from the current sample were below the cut-off values for dysphonia. Mean HNR for the prolonged vowel was above 20 dB, indicating a normal voice.
Comparison of cepstral peak prominence measures using the ADSV, SpeechTool, and VoiceSauce acoustic analysis programs in vocally healthy female speakers.
Comparison of cepstral peak prominence measures using the ADSV, SpeechTool, and VoiceSauce acoustic analysis programs in vocally healthy female speakers.
Cohen's d and P-values are based on (t test) comparisons between the musically trained and non-trained groups. References (Ref) are provided to alert to normative data comparisons where available (NA, not available).
(*) significant at P < 0.05.
Abbreviations: CPPS, smoothed cepstral peak prominence; CSID, cepstral spectral index of dysphonia; F0, fundamental frequency; HNR, harmonics-to-noise ratio; SD, standard deviation.
The musically trained group had significantly lower CSID of vowel, t(112) = 2.49, P = 0.01, d = 0.47 and CSID of the 3rd CAPE-V phrase, t(112) = 2.34, P = 0.02, d = 0.44, in comparison to the non-trained group as shown in Table 2. No other acoustic voice measures distinguished the subgroups.
Instrumentalist and vocalist groups
Comparisons across the instrumentalist and vocalist subgroups are shown in Table 3. The vocalist group had significantly lower vowel F0SD, t(56) = 2.05, P = 0.05, d = 0.55 in comparison to the non-trained group. No other voice production measures differentiated the subgroups.
TABLE 3Sample Means (SD) for Voice Data for the Instrumentalist and Vocalist Groups
Comparison of cepstral peak prominence measures using the ADSV, SpeechTool, and VoiceSauce acoustic analysis programs in vocally healthy female speakers.
Comparison of cepstral peak prominence measures using the ADSV, SpeechTool, and VoiceSauce acoustic analysis programs in vocally healthy female speakers.
Cohen's d and P-values are based on (t test) comparisons between the instrumentalist and vocalist groups. References (Ref) are provided to alert to normative data comparisons where available (NA, not available).
(*) significant at P ≤ 0.05.
Abbreviations: CPPS, smoothed cepstral peak prominence; CSID, cepstral spectral index of dysphonia; F0, fundamental frequency; HNR, harmonics-to-noise ratio; NA, not available; SD, standard deviation.
Relationships between pitch discrimination accuracy and acoustic voice measures
Correlations between the measures were first performed on the whole sample as detailed on the left of Table 4. No voice quality measures showed correlations with PD accuracy across the whole cohort (all r values < 0.2).
TABLE 4Pearson's Correlation Coefficients (r) for the Pitch Discrimination and Voice Data for the Whole Sample, Musically Trained and Non-Musically Trained Groups and Healthy Voice and Mildly Voice Disordered Groups
Tasks
Acoustic Measures
Whole Sample (n = 114)
Musical Trained Group (n = 58)
Non-Musical Trained Group (n = 56)
Instrumen-talists (n = 48)
Vocalists (n = 10)
Rainbow passage
F0 (Hz)
-0.10
0.11
-0.19
0.12
0.35
CPPS (dB)
-0.08
0.13
-0.33*
0.12
0.38
CSID
0.10
-0.17
0.15
-0.20
0.21
Intensity (dB)
-0.11
0.14
-0.32
0.17
0.46
3rd CAPE-V phrase
F0 (Hz)
-0.20
-0.09
-0.26
-0.11
0.16
CPPS (dB)
0.03
0.16
-0.06
0.14
0.30
CSID
-0.11
-0.13
0.09
-0.09
-0.51
Intensity (dB)
-0.06
0.15
-0.24
0.19
0.30
/a/ vowel
F0SD (Hz)
-0.18
-0.37*
0.01
-0.39*
0.29
HNR (dB)
-0.07
0.02
-0.29
0.02
-0.05
CPPS (dB)
0.05
-0.06
0.13
-0.05
0.11
CSID
-0.18
-0.10
-0.08
-0.15
0.24
Intensity (dB)
-0.02
0.04
-0.20
0.04
0.36
All measures are in dB, unless otherwise stated.
(*), Bonferroni-adjusted P < 0.05.
Abbreviations: CPPS, smoothed cepstral peak prominence; CSID, cepstral spectral index of dysphonia; F0, fundamental frequency; HNR, harmonics-to-noise ratio; SD, standard deviation.
Within group relationships for the musically trained and untrained speakers
In the musically trained group, a small negative correlation was observed between pitch accuracy and vowel F0SD only, r = -0.37, P = 0.004, with pitch accuracy explaining about 14% of the variance in this measure. Greater PD was associated with low variance in fundamental frequency production. We have plotted data for the whole sample in Figure 3, with the individuals with a musical background shown as solid symbols. No other acoustic measures showed correlations with PD accuracy (all r values < 0.3).
FIGURE 3Scatter plot depicting the relationship between perceptual discrimination accuracy (%) and F0SD voice quality for musically trained (solid symbols/regression line) and non-trained participants (open symbols/dashed regression line).
In the non-trained subgroup (n = 56), small negative correlations were observed between pitch accuracy and Rainbow Passage Intensity (r = -0.32, P = 0.02) and Rainbow Passage CPPS (r = -0.33, P = 0.01). Better PD was also associated with worse performance on the vowel HNR voice measures (r = -0.29, P = 0.03), although this correlation was not significant at the adjusted p level. Other acoustic measures did not show correlations with pitch accuracy (all r values < 0.30).
In the non-trained subgroup, vocal intensity in the Rainbow Passage was correlated with both Rainbow Passage CPPS, r = 0.61 and Vowel HNR, r = 0.46 (both Ps < 0.001). No correlations between vocal intensity, CPPS and HNR were observed in the mildly voice disordered and non-disordered subgroups.
Within group relationships for the instrumentalist and vocalist groups
Among the relatively large group identified as instrumentalists, a small correlation was observed between PD accuracy and F0SD (r = -0.39, P = 0.006). Among the vocalists, there were a number of correlations of 0.30 or greater between PD accuracy and acoustic voice measures, but due to the small sample size, none of these were statistically significant. Of most note was the correlation of -0.51 between PD accuracy and CAPE-V3 CSID, meaning that better PD accuracy was associated with better voice quality, however this correlation was not statistically significant (Table 4).
DISCUSSION
There were no statistically significant relationships between any of our various measures of voice quality and pitch perception among our sample of 114 female participants who self-reported as having a non-disordered voice. Individual differences in perception of pitch were not concomitantly manifest in individual differences in voice production. Although musicians had generally better pitch perception than non-musicians, there was little evidence for a relationship between production and perception. Moreover, the musicians only differed from the non-musicians in one out of 13 acoustic voice measures.
Validity of NeAP as a pitch discrimination testing tool
The NeAP was shown to be a reliable tool in testing PD ability. It showed moderate to good prediction accuracy in differentiating musical training background. Whilst the PD accuracy in this study was similar to other research in non-disordered speakers,
further studies to establish the best protocol to use in terms of number of tone pairs and size of tone difference should also be explored. It is possible that with measures of just noticeable difference, there may be increased variance and sensitivity to pitch differences.
Pitch discrimination accuracy and voice characteristics
The whole cohort had a PD accuracy of approximately 70%. Currently, no large cohort baseline data exists on non-clinical populations for voice, so these data on pitch perception will help to establish such norms. This data is comparable to the criteria of 75% discrimination accuracy used to distinguish participants who ‘accurately discriminated’ tone pairs in a previous study of singers, musicians, and non-musicians.
Same or different pitch? Effects of musical expertise, pitch difference, and auditory task on the pitch discrimination ability of musicians and non-musicians.
musical training background had a strong effect on PD accuracy, even though none of the participants identified as professional musicians. Participants with musical training had better PD accuracy than those without. In the musically trained subgroup, PD accuracy was only correlated with 1 out of 13 acoustic voice measures. Higher PD accuracy was weakly correlated with lower F0SD, which is indicative of better control of F0 in a prolonged vowel task. This general absence of a relationship between PD and acoustic voice data may be a consequence of the fact that participants had low, variable and for the most part, early development musical training. Indeed, positive correlations were observed between PD ability and pitch matching accuracy in musicians in a previous study.
The use of different measures of PD and vocal production across studies might partially explain the lack of any significant relationship.
We used a range of acoustic voice measures, including both frequency-based and spectral-based, to represent overall vocal function/production. Whilst the frequency measures (F0 and F0SD) were more relevant to the PD task, other acoustic measures reflect the overall voice quality (ie, CPPS and CSID). The control of voice production requires more than just the control of pitch. For participants with musical training experience, the auditory system would likely have been trained to attend to pitch, such as when tuning an instrument. As such, there is reason to think it would also be more sensitive to minor pitch changes in voice. For those without a musical background, the mechanism of voice control may not be driven by factors related to pitch. Therefore, the use of overall voice quality measures was reasonable. Our results seemed to suggest that conscious, auditory perception might play a weak role compared with other sensory feedback modalities such as proprioception, or non-conscious perceptual processes as suggested in the Linked Dual Representation Model [23], in regulation of vocal production.
In the musically untrained subgroup, PD accuracy was weakly correlated with 1 out of 13 acoustic voice measures and another two vocal production measures showed trends of correlation with PD accuracy. Individuals with better PD accuracy were found to havea lower CPPS of the Rainbow Passage and lower HNR in the prolonged vowel task, indicative of a poorer voice quality. Existing evidence at most shows a dissociation between vocal perception and production,
not a negative relationship. It may be that the lower CPPS values were caused by reduced vocal intensity. In the non-trained group, PD score showed a trend to be correlated with vocal intensity of the Rainbow Passage (r = -0.32, Table 4). This is consistent with existing evidence that shows lower vocal intensity is correlated with lower CPPS and HNR values
In the non-trained subgroup, vocal intensity in the Rainbow Passage was correlated with both Rainbow Passage CPPS, r = 0.61 and Vowel HNR, r = 0.46 (both Ps < 0.001). No correlations between vocal intensity, CPPS and HNR were observed in the mildly voice disordered and non-disordered subgroups.
As such, it is likely that these weak associations were a consequence of using a softer voice rather than a degraded voice quality. It is possible that untrained speakers use a different mechanism to control their voices compared to musically trained people, monitoring their vocal loudness rather than pitch. Indeed, in a typical speaker, vocal intensity is an important feature that ensures communicative effectiveness, especially when speaking in an environment with ambient noise.
It has been shown experimentally that when speakers are trained to produce a novel vocal task, individuals make significant changes to reduce vocal intensity to achieve improved vocal productions.
Show and tell: video modeling and instruction without feedback improves performance but is not sufficient for retention of a complex voice motor skill.
vocalists had significantly better PD abilities and more stable fundamental frequency control than instrumentalists (although there were only n = 10 vocalists causing statistical issues in variance comparing across vocalists and instrumentalists, of which the latter group had n = 48). Whilst vocalists did not demonstrate better vocal production abilities in other acoustic measures, better PD supports the idea that training effects are more specific rather than general. For participants with vocal training experience, the vocal production system would likely have been trained to match or hold a stable pitch when singing, therefore, it is likely that vocalists would be more sensitive to minor pitch changes. As such, our results suggest that vocal production in specific tasks may play a small role in regulating auditory perception abilities.
PD accuracy only showed a trend to be correlated with more stable fundamental frequency control (lower F0SD) in instrumentalists. No statistically significant correlations were observed between PD accuracy and vocal production measures in vocalists (although there were small to medium sized effects noted for a number of measures and all but three of the correlations were greater than r = 0.20, Table 4). The lack of significant correlations observed in this study is consistent with results in highly trained, working vocalists,
and may be due to low variability in both PD abilities and vocal production in this group.
With the extremely small sample of vocalists (n = 10), any conclusions based on comparisons or correlations should be treated cautiously. In future work it will be important to consider a larger cohort of vocalists to understand the relationship between vocal perception and production abilities; ideally performing developmental work to determine if and how these abilities co-develop.
Implications for theoretical models of perception and production
The findings can be used to understand some theoretical models related to perception and production. According to the Theory of Event Coding,
we would expect correlations between perception and production in a non-disordered population of speakers. However, PD was not significantly correlated with any of the vocal production measures in the general cohort. Correlations with PD accuracy were only observed in participants with musical training background and only in measures closely associated to specific skills (ie, pitch production) that is important in musical training.
model, pitch perception and vocal production can be modulated in two separate pathways. In one pathway, vocal production can be mediated by conscious perceptual judgments of sound, whereby some sort of perceptual trace or representation guides action production.
Accordingly, those with better pitch perception abilities will often also have better pitch production abilities, but the reverse relationship does not hold. Our data is not consistent with this uni-directional relationship between perception and production. We did not observe a relationship between PD and any acoustic voice measures in the whole cohort. In addition, it was the vocalists who had more stable vocal production skills that were linked to better PD, compared to instrumentalists. This vocalist-instrumentalist discrepancy suggests a bi-directional relationship between voice production and PD. Alternatively, people with naturally stable phonation may tend to become vocalists rather than instrumentalists. They may also have better pre training PD compared to instrumentalists and musically untrained people; however, this should be investigated in future studies.
CONCLUSION
The relationship between auditory perception and vocal production continues to be elusive, yet important for theoretical and practical reasons. Based on multiple measures of acoustic voice quality in this study, it would appear that in a population of individuals who would be considered clinically non-voice disordered, with low musical training, there is little to no relationship between PD and voice quality. Where relationships were observed, these were suggestive of a uni-directional relationship whereby conscious perception influences vocal production in individuals with some musical training, consistent with the Linked Dual Representation
model. These perceptual to production transfer benefits were limited to measures specific to pitch, in this case fine control of fundamental frequency. These results support the idea that training the perceptual system will influence the control of the production system in the specific domain in which it is trained (eg, frequency, intensity or spectral features). However, the reverse relationship was not observed in this study. Although vocalists had improved pitch perception skills compared to instrumentalists, we did not observe statistically significant correlations between perception and production measures (likely because we were underpowered to detect such effects). These results suggest that we cannot rule out a bi-directional relationship between perception and production. Our data also provide preliminary evidence that individuals without musical training might monitor their vocal production through the perception of intensity (ie, loudness). Future studies must therefore ensure that the PD and acoustic voice measures used are appropriate to the skills that are being investigated.
Author contribution statements
Emily Wing-Tung Yun conducted the literature review, prepared the research protocol, recruited participants, collected data, performed acoustic voice analyses and wrote the manuscript. Duy Duong Nguyen was involved in research question identification, data collection, data analysis, data interpretation, manuscript writing and editing, and graphic works. Paul Carding and Nicola Hodges were involved with reviewing and editing the manuscript. Robert Heard assisted with data interpretation and analysis. Antonia Chacon was involved in recruitment of participants and data collection. Catherine Madill conceived the research idea, wrote and edited the manuscript. All authors reviewed and approved the final version of the manuscript.
Competing interests
The authors have no competing interests to declare in this study.
Acknowledgments
We'd like to acknowledge the contribution of Dr. Robert Heard at The University of Sydney for his initial advice regarding statistical analysis and research design.
APPENDIX 1. Reference and Comparison Tones Used in the Default NeAP PD Task
Same or different pitch? Effects of musical expertise, pitch difference, and auditory task on the pitch discrimination ability of musicians and non-musicians.
Comparison of cepstral peak prominence measures using the ADSV, SpeechTool, and VoiceSauce acoustic analysis programs in vocally healthy female speakers.
Show and tell: video modeling and instruction without feedback improves performance but is not sufficient for retention of a complex voice motor skill.
00054-6&id=gr1.jpg){kind=link}
00054-6&id=gr2.jpg){kind=link}
00054-6&id=gr3.jpg){kind=link}