Identification of Stop-Nasal Cognates as a Function of Increasing Velopharyngeal Coupling in Children

Abstract

Objective

To investigate the influence of velopharyngeal (VP) port area on the perception of oral stop versus nasal consonants in child-sized vocal tracts.

Design

C₁VC₂ words were generated using a computational model scaled to 4-, 6-, and 8-year-old vocal tracts. VP coupling area was systematically varied along a continuum from 0 to 0.1 cm². Stimulus presentation was randomized and presented in a forced-choice perceptual identification task with 4 response options reflecting consonant differences.

Setting

University laboratory.

Participants

Thirty untrained adult listeners.

Intervention

Incremental increases in VP coupling area.

Main Outcome Measures

Listener identification of oral versus nasal consonants.

Results

Across all vocal tract ages and vowel contexts (/æ, ɪ, u/), crossover areas at which listeners perceived a nasal instead of an oral stop ranged from 0.014 to 0.035 cm². A one-way multivariate analysis of variance revealed a significant effect of talker age on overall vowel measures, with post hoc tests showing 4-year-olds had smaller crossover areas than older children for /æ/. Crossover areas for child vocal tracts were smaller than those reported for adults (∼0.045-0.046 cm²). Variability across vowels and ages, particularly in younger talkers, indicated broader perceptual ambiguity.

Conclusions

Small VP openings in child vocal tracts can produce perceptual cues typically associated with hypernasality, even at relatively modest gap sizes. These findings reflect developmental differences in speech production and may inform clinical management of VP insufficiency, emphasizing the perceptual impact of VP gap size on listener judgments.

Introduction

Children with velopharyngeal (VP) dysfunction exhibit speech difficulties that significantly reduce intelligibility due to excessive nasal coupling. Nasal coupling occurs when the VP port is open, permitting acoustic coupling of the oral and nasal cavities during the production of nasal phonemes (/m/, /n/, /ŋ/). The VP mechanism typically enables the production of oral phonemes by separating the oral and nasal cavities. In VP dysfunction, however, the port remains open during oral phoneme production when closure is required, resulting in inappropriate nasal coupling. Consequences resulting from VP dysfunction include hypernasality, weak pressure consonants, and audible nasal air emission. Variations in VP port area influence the degree of nasal resonance, with larger openings yielding greater nasal influence and smaller yielding less. ¹ The bulk of this early research quantifying VP port area was first completed using aerodynamic assessment in children and adults with repaired cleft palate by Warren et al.^2–5

Direct visualization of the VP mechanism is commonly achieved clinically through nasopharyngoscopy. Perceptual ratings of nasality, which are primarily based on vowels,^6,7 are frequently confirmed via visualizing the VP mechanism to determine its presence, cause, and potential treatment of VP dysfunction. Dalston et al ⁸ found weak correlations between VP gap size and measures of nasalance, perceptual resonance ratings, and aerodynamic estimates of VP orifice size. Kummer et al ⁹ studied the relationship between perceptual characteristics of hypernasality, nasal emission, and nasal rustle with VP gap size; clinicians correctly predicted VP gap size from speech characteristics in approximately 70% of cases, with moderate and severe hypernasality associated with predictions of larger gaps and nasal rustle associated with smaller gaps. However, Youssef and Alkhaja ¹⁰ also reported weak associations between perceptual judgments and gap size.

Although nasopharyngoscopy provides valuable real-time visualization of VP function during speech, quantifying VP gap size from endoscopic images remains inherently challenging. Variability in scope positioning, viewing angle, and clinical interpretation can affect estimates of gap size, limiting the precision with which structural observations can be directly related to acoustic output and perceptual judgments. While this variability can be due to a variety of factors, including clinician skillset, these challenges highlight the ongoing need for approaches that allow systematic examination of specific VP parameters under controlled conditions.

Studying the impact of VP gap size on perceptual ratings in participants with VP dysfunction poses challenges, one of which is the inability to control VP gap size while holding other factors, such as articulation, constant. Various types of speech synthesizers have been used to address this challenge since they are used to generate sets of acoustic signals presented to listeners based on incremental parameter variations such as VP gap size. House and Stevens, ¹¹ House, ¹² and Hecker ¹³ all used electrical circuit analogs of the vocal tract and nasal tract to synthesize stimuli for perceptual experiments in which listeners responded to some aspect of nasalization of vowels and consonants based on varying the VP coupling area (see also review by Curtis ¹⁴ ). The range of area values indicated in these studies that affected responses for nasal categories were much larger than values later reported in studies that used more complex computer-based articulatory synthesizers.^15,16 This was likely because the circuit-based vocal tract models had none of the built-in constraints on articulation that were included in the articulatory synthesizer. More recently developed computational models representative of the function and structure of speech production have been utilized to synthesize speech sounds for which many parameters can be controlled. Through this approach, the relation of systematic parametric variations, such as the exact size of the VP opening, to both acoustic characteristics and listener perception of that same signal can be studied (cf Bunton and Story ¹⁷ ). Several studies using simulated adult speech have incrementally altered the degree of nasal coupling in vowels and have shown that both nasalance scores and listener ratings of nasality increase as VP port size increases. ¹⁷

Computational modeling has also been used to investigate the effect of varying degrees of nasal coupling on the identification of stop versus nasal consonants. Using a speech production system representative of an adult male talker, Story and Bunton ¹⁸ presented continua of VCV utterances to listeners for which the maximum VP coupling area for the consonant varied from 0 to 0.1 cm². Depending on the place of consonantal constriction and vowel context, the maximum VP port area at which listeners transition from identifying a consonant as a stop to a nasal ranged from 0.035 to 0.057 cm². A similar study was reported by Story and Bunton ¹⁹ where, in addition to the simulated adult male talker, stimuli were generated based on scaling the speech production system to be representative of an adult female talker and 4-year-old child talker. The scaling was applied to the length and cross-sectional areas of the vocal tract and trachea, but for the nasal tract, only the length was scaled (ie, the cross-sectional areas of the nasal passages and coupled sinuses were the same regardless of talker size). In addition, the location of the VP port was set to be halfway between the glottis and lips, a fairly accurate location for the adult male system, but perhaps less accurate for the smaller female and child-like systems. The laryngeal system was also scaled for the adult female and 4-year-old child simulation such that the fundamental frequency and flow-generating mechanisms were appropriate for the system size. Results indicated that scaling of the vocal tract and vocal fold parameters had little to no impact on the VP coupling area where listeners made this perceptual transition. The range of coupling areas at which consonants were identified as nasals was 0.037 to 0.049 cm² for the male model, 0.040 to 0.055 cm² for the female model, and 0.039 to 0.052 cm² for the 4-year-old child model. ¹⁹ Comparisons of these listener-based ranges to 2 types of nasalance calculations were well aligned for the adult male simulated stimuli, but somewhat less so for the female and child-like system. Further, the nasalance calculated for the child-like stimuli suggested that listeners would identify nasal consonants at a smaller VP opening area than was observed.

The purpose of this study was to determine the size of the nasal coupling area (ie, VP gap size) at which listeners switch from identifying consonant manner as oral stop versus nasal based on C₁VC₂ words generated using a simulated child-size speech production model representative of 4, 6, and 8 years of age. In contrast to Story and Bunton, ¹⁹ in this study the length and cross-sectional areas of the vocal tract, trachea, and nasal tract (passages and sinuses) were scaled for each simulated talker, and the VP port location was more accurately set to be at 40% of the distance between the glottis and lips. Based on previous literature, it was hypothesized that the area threshold would be similar to that reported for syllables produced by adult talkers, ¹⁸ but that younger models would show a more gradual trend of listener responses and a greater standard deviation over which consonants were identified as stops versus nasals.

Method

The institutional review board at the University of Wyoming approved all procedures, and a reliance agreement was in place with the University of Arizona. Informed consent was obtained from all participants.

Brief Overview of Speech Production Model

Stimuli for this study were generated by a speech production model that includes tubular representations of the trachea, vocal tract, and nasal tract, as well as a kinematic representation of the vocal folds.^18–21 The components of the model are shown schematically in Figure 1 and include the vocal tract, piriform sinuses, velopharynx, nasal passages, sphenoid sinus, and maxillary sinuses; the trachea is included in the model but is not indicated in the figure. The schematic in Figure 1 is specifically configured to represent an 8-year-old child talker where the overall vocal tract length (VTL) is 12.65 cm based on Story et al, ²² and the location of VP coupling is 5.18 cm from the glottis. ²² All airway components are represented in the model as an assembly of cross-sectional areas (area function) and function as a complex acoustic waveguide.

Figure 1.

Schematic of the child-size (8 year old) speech production model. The x-axis is anterior–posterior (cm) and the y-axis is inferior–superior (cm). The filled arch represents the vocal tract. The nasal tract extends above and terminates at the nares (emitting relatively less acoustic energy), while the vocal tract terminates at the lips (emitting relatively more acoustic energy).

The vocal tract is shown in Figure 1 as a pseudo-midsagittal representation of 44 cross-sectional areas, including the epilarynx, pharynx, and oral cavity in a neutral vowel configuration that radiates sound at the lips. Based on Story et al, ²² each of the 44 sections for the 8-year-old child-like model had a length of 0.2877 cm for a total VTL of 12.65 cm as mentioned in the previous paragraph. ²² The section lengths for the 6-year-old and 4-year-old models were set to 0.2743 cm and 0.2586 cm, respectively, generating VTLs of 12.07 cm and 11.38 cm. The cross-sectional areas were also scaled based on the age-dependent model presented in Story et al. ²² When the VP coupling area is nonzero (ie, at least partially open), the nasal passages and sinuses are integrated into the overall model and radiate an acoustic signal at the nares. The configuration of the nasal passages and sinuses is based on data reported by Story ²³ and Pruthi ²⁴ for an adult male but the section lengths were scaled to be identical to those of the main vocal tract for each age and cross-sectional areas were scaled by factors of 0.42, 0.48, and 0.53, respectively for the 4-, 6-, and 8-year-old models. Coupling of the nasal tract to the vocal tract results from lowering the velum and has the effect of slightly modifying the shape of the vocal tract in the velar region. This change was accounted for by adjusting the VP portion and corresponding region of the vocal tract area function based on the “distributed coupling method” explained by Pruthi et al, ²⁴ and much like the technique used by Feng and Castelli ²⁵ and Maeda. ²⁶

The first step in producing a stimulus with the speech production model is to generate the time-variation of the vocal tract area function (ie, shape) for a target utterance. This is done by representing each phonetic segment of the target utterance by an acoustic target that prescribes the upward or downward deflections of the vocal tract resonance frequencies relative to those of either a neutral tract configuration in the case of vowels, or relative to an underlying vowel in the case of consonants (see Story and Bunton for details^18,19,21). Each acoustic target is associated with an “event function” that prescribes the time course of the vocal tract modulation that results from any given acoustic target. Shown in the lower panel of Figure 2 are 3 event functions that represent the temporal durations of /b/, /æ/, and /d/ to produce “bad”; the schwa at the beginning of the utterance is generated without a specified target and event since the “resting state” of the vocal tract is a neutral configuration.

Figure 2.

Speech production model output for “a bad” to “a ban.” Top: waveform produced when max(A_n(t)) = 0.1 cm² along with phonetic transcription; red portion indicates sound radiated at the nares. Middle: 11 time-varying velopharyngeal coupling functions (A_n(t), cm²). Bottom: event functions controlling vocal tract movements; “d/n” indicates nasalization of the oral stop due to VP coupling.

The acoustic targets and event functions generate a time-varying vocal tract area function that represents the articulatory actions to produce a given utterance—in this case “a bad.” Shown in the top panel of Figure 2 is a collection of time-dependent VP coupling functions, A_n(t), that are time-synchronous with the event functions and dictate the time course of the VP coupling during production of the utterance. With the exception of the “no-nasal” function that is equal to 0.0 throughout the entire duration, all other A_n(t) functions impose some amount of coupling of the nasal system to the vocal tract during the time interval from about 0.3 s to the end of the utterance; when the VP coupling area is sufficiently large, the utterance will become “a ban.”

Time-varying vocal tract area functions were generated for representative child talkers at ages 4, 6, and 8 years of age producing “bad,” “bid,” and “bud” and then in combination with 3 sets of 11 VP coupling functions, where one set affected the initial consonant, a second set affected the final consonant, and the third set affected both consonants. In total, 297 combinations of vocal tract modulations and VP coupling functions were generated.

Once a time-varying area function has been generated, the speech production model generates an audio signal by coupling an interactive voice source model to the tracheal exit and vocal tract entry. At each time sample, acoustic wave propagation in the airways is calculated over the time course of the utterance with a wave-reflection algorithm^23,27,28 that includes energy losses due to yielding walls, viscosity, heat conduction, and radiation at the lips, nares, and skin surfaces (specifically as described in Story ²³ ). The acoustic output signal is the sum of the radiated pressure at the lips, nares, and skin surfaces. The waveform in the top panel of Figure 2 was generated by the model when the A_n(t) had a maximum value of 0.1 cm², where the red portion of the waveform indicates the sound radiated at the nares. Similar collections of VP coupling functions were generated to coincide with the initial consonant /b/ alone, as well as both the initial and final consonants, thus setting up a continuum where /b/ shifts to /m/ and /d/ shifts to /n/.

Simulation of Audio Stimuli

Stimuli consisted of C₁VC₂ words embedded in the carrier phrase “a ___.” The CVC format was selected to capture coarticulatory influences of bilabial and alveolar consonants on adjacent vowels while minimizing unwanted interactions. ²⁹ Stimuli were designed to transition perceptually from stop consonants to nasals, with VP port opening size systematically varied to simulate this shift. ¹⁸

Acoustic stimuli, 0.6 s in duration, were generated for the 3 simulated talker ages (4, 6, and 8 years of age), based on the model described in the previous section. Source characteristics, including fundamental frequency, were consistent with those expected for children at these ages. The fundamental frequency (f_o) contour started at 80% percent of the maximum value then increased to the maximum value at 0.25 s then decreased to 65% of the maximum by the end of the utterance; the maximum f_o values were 318 Hz, 295 Hz, and 272 Hz, for the 4-, 6-, and 8-year-old models, respectively. Stimuli amplitudes were normalized to 80% of the dynamic range to ensure consistent presentation levels across conditions.

Nasal Consonant Identification Experiment

Stimuli

CVC words with coarticulatory structures were generated and presented to listeners. Each stimulus varied by consonant (/b, d, m, n/) in either the initial or final word position and vowel (/æ, ɪ, u/) context, producing sets of 4 possible word choices (eg, for /æ/: bad, ban, mad, man). Stimuli were chosen to maintain consistent place of articulation across conditions, and foils were restricted to real words to avoid confounding effects of lexical status on listener judgments. The VP coupling area was systematically varied across an 11-step continuum. ¹⁸ This created the 297 total audio samples to be played to listeners (3 ages × 3 vowel contexts × 3 consonant manipulations [initial, final, or both] × 11-step continuum).

Participants

Thirty naïve listeners (Mean_age = 23.43 years; Range_age = 19-49 years; equal number of males and females) were recruited from the University of Wyoming campus through flyers. All reported being native speakers of American English with no history of speech, language, or hearing disorders. Participants were excluded if they failed a hearing screening at 25 dB HL across 500, 1000, 2000, and 4000 Hz. Each listener was randomly assigned to hear 2 age conditions (4-, 6-, or 8-year-old versions of the model), with stimuli presented across all 3 vowel contexts. All listeners took a mandatory break between the 2 age conditions to mitigate listener fatigue. This resulted in 20 listeners per stimulus and age. Due to inconsistencies in the data, one age/vowel combination (/æ/ for 4-year-old talker) had only 19 listeners.

Procedure

Listeners were seated in a sound booth and presented stimuli binaurally over headphones (Sennheiser HD 206). The Alvin interface ³⁰ controlled stimulus presentation and response collection. Three experimental trials (one for each age condition) were created, with order randomized across listeners. Samples were blocked by age and vowel. Before the experiment, participants completed a short training session, which consisted of 12 practice identifications identical to the perceptual experiment tasks that did not count toward the results.

During the task, listeners completed a forced-choice paradigm: 4 possible word options appeared on screen, and participants used the mouse to select the word that best matched what they heard. Each of the 99 unique stimuli for a given talker (11 step continuum × 3 vowels × 3 consonants manipulations) was presented 5 times in a randomized order, and listeners could replay tokens as needed. In total, each participant provided 990 responses (99 stimuli × 2 ages × 5 presentations).

Data Analysis

Responses were categorized as “stop” or “nasal” based on the consonant identified, relative to the manipulated VP coupling area. Identification data were normalized and fit with a curve using MATLAB's pchip algorithm.^18,31 The perceptual shift from stop to nasal was defined as the VP coupling area at which nasal responses exceeded 50%.^18,19 As examples, listener identification plots are shown in Figure 3 for 2 continua produced by the 6-year-old model talker, where Figure 3a is the continuum from “bad” to “ban” and Figure 3b is “bud” to “mud.” The solid black line (with associated black dots as data points) in Figure 3a indicates the percentage of responses for each stimulus along the VP area continuum for which listeners chose “bad” as the utterance; the dotted orange line indicates the “ban” responses. The VP area at which the responses cross-over from stop to nasal is indicated by the vertical line and is 0.035 cm² in this case. Figure 3b shows listener identification plots for the “bud” to “mud” continuum, where the solid black line indicates the percentage of listener responses for “bud” and the dotted orange line tracks responses for “mud.” In this case, the VP area at which the stop shifts to a nasal is 0.014 cm².

Figure 3.

Example listener identification curves as a function of increasing velopharyngeal opening for /æ/ vowel produced by the 6-year-old model talker. The vertical line indicates the VP area at which the responses switched from stop to nasal consonant. (a) “bad” to “ban” continuum, and (b) “bud” to “mud” continuum.

Statistical Analysis

Response categories were collapsed to capture only change related to the phoneme that was impacted by VP opening (ie, initial consonant). Any response of /b/ or /d/ was labeled as a stop while any response of /m/ or /n/ was labeled as a nasal. A one-way multivariate analysis of covariance was used to assess differences in consonant identification by vowel type while controlling for speaker age. Descriptive statistics were provided for each age range. Each sample was repeated 3 times to calculate intrarater reliability. An intraclass correlation coefficient was used to calculate reliability. Intrarater reliability was assessed using a 2-way mixed-effects intraclass correlation coefficient (ICC[3,k]) across the 3 repetitions per listener. Results indicated moderate-to-high reliability (ICC = 0.82, 95% CI [0.77-0.86]), suggesting that listeners were generally consistent in their judgments across trials.

Results

Crossover areas for oral stop versus nasal consonant identification ranged from 0.014 to 0.035 cm² across all vowels and ages. A one-way multivariate analysis of variance (MANOVA) was conducted to examine age-related differences across the 3 vowel measures. Box's M test indicated that the assumption of equality of covariance matrices was violated (Box's M = 75.89, F (18,70964) = 4.09, P < .001). Given roughly equal group sizes, Pillai's Trace was used to interpret multivariate effects, as it is robust to this violation. There was a statistically significant difference based on the overall combination of dependent variables after controlling for talker age, V = 0.149, F (6352) = 3.585, P < .001, partial η² = 0.050.

For 4-year-old talkers, crossover areas varied across vowel contexts, with mean crossover areas largest for /æ/, followed by /u/ and /ɪ/ (Table 1). Post hoc comparisons indicated that for the /æ/ vowel, 4-year-olds exhibited significantly smaller crossover areas than 8-year-olds (P = .013). For 6-year-old talkers, crossover areas were similar across vowel contexts, with no significant age-related differences. For 8-year-old talkers, crossover areas tended to be largest for /æ/ and smallest for /u/.

Table 1.

Mean Crossover Areas (cm²) for Stop–Nasal Identification by Age and Vowel, With Significant Post Hoc Age Differences.

Age	Vowel	Mean ± SD	Significant Age Difference*
4	æ	0.024 ± 0.008	vs. 8
	ɪ	0.019 ± 0.012	–
	u	0.020 ± 0.013	vs. 8
6	æ	0.028 ± 0.008	–
	ɪ	0.023 ± 0.010	–
	u	0.018 ± 0.008	–
8	æ	0.029 ± 0.011	–
	ɪ	0.022 ± 0.013	–
	u	0.016 ± 0.008	–

*Significant age differences are shown only for post hoc contrasts that reached P < .05. Direction of effects is reflected in the group means.

Follow-up univariate analyses revealed a significant effect of age for the /ɪ/ vowel, F (3206) = 5.22, P = .002, partial η² = .071, whereas age effects were not significant for /æ/ or /u/ (ps > .05).

Discussion

This perceptual study examined how increasing nasal coupling area affects consonant perception in children using a computational model scaled to 4-, 6-, and 8-year-old vocal tracts. C₁VC₂ words were generated with incremental increases in nasal coupling areas ranging from 0 to 0.1 cm². Crossover areas for oral stop consonant versus nasal consonant identification ranged from 0.014 to 0.035 cm² across all vowels and vocal tract ages. A one-way MANOVA revealed a significant effect of age, with post hoc tests showing that 4-year-olds had smaller crossover areas than older children for /æ/. A univariate effect of age was also observed for /ɪ/, although effects were not significant for /u/, highlighting that age-related perceptual differences may vary across vowel contexts. It is important to note that these findings are based on computational simulations rather than actual child speech, which limits generalization to natural production.

These crossover areas were smaller than those reported for a simulated adult male talker (0.035-0.057 cm²), ¹⁸ and also smaller than reported for a simulated 4-year-old talker (0.039-0.052 cm²). ¹⁹ The smaller crossover area may, at least in part, have been affected by the additional scaling of the nasal and the difference in VP coupling location, both of which were different in Story and Bunton ¹⁹ (ie, the nasal tract areas were not scaled and coupling location was exactly halfway between glottis and lips). It is also not clear what effect the CVC context, rather than VCV as in Story and Bunton,^18,19 may have had on the results. Additionally, past studies of VP opening during rest breathing and /m/ report larger areas than those reported in the present study and likely represent near maximal VP opening. Sundström and Oren ³² reported VP openings ranging from 0.02 cm² in a child who had been diagnosed with severe nasal emission (normal resonance) to 0.44 cm² in a child with hypernasality. These data provide a useful range of VP coupling areas corresponding to turbulence generation at the low end (eg, with the relatively high oral pressure during a /z/ and extensive nasalization at the high end). However, the crossover point range in the present study is closer to the VP opening size previously reported for nasal turbulence. Nasal turbulence, however, was not considered in this model, so it is unclear how that may impact our results.

This study demonstrates that small VP openings can significantly affect consonant perception, meaning that even modest gaps can produce perceptual cues commonly associated with severe hypernasality, despite the actual VP opening being relatively small. Clinically, severe hypernasality is often associated with large VP gaps, whereas smaller VP gaps typically result in audible nasal emission. In the Cleft Audit Protocol for Speech-Augmented (CAPS-A) ³³ and CAPS-A-Americleft Protocol (CAPS-A-AM), ³⁴ 2 commonly used speech assessment protocols for cleft palate, speech-language pathologists rate hypernasality on a scale from 0 to 4 (0 = normal, 4 = severe). According to these protocols, only when hypernasality is rated “severe” are oral consonants impacted. This study demonstrates that characteristics commonly associated with severe hypernasality can result from relatively small gap sizes. Further, a patient who exhibits consistent hypernasality typically undergoes instrumental assessment to obtain additional information about cause and VP gap size. ³⁵ In contrast to those with relatively moderate to large VP gap sizes, those with smaller VP gap sizes may be managed more conservatively, often not being candidates for surgical intervention or treated with fat injections of variable efficacy.^36,37 Findings from this study support patient-specific management and treatment based on speech impact at the individual level rather than generalized protocols based on VP gap size alone.

VCV stimuli produced by the adult-like speech production models required an VP coupling area of about ∼0.045 cm² for listeners to perceive a nasal consonant,^18,19 whereas the CVC stimuli generated by all 3 child-like models in the present study reached the stop–nasal identification cross over at smaller coupling areas. Variability across vowels and ages, particularly in 4-year-olds, supports the hypothesis that younger speakers produce more perceptually ambiguous acoustic cues at smaller VP gaps. The developmental trend toward larger crossover areas with age may reflect maturational changes in articulatory control and acoustic output. As the size of the pharynx increases, smaller gaps may be less impactful on consonant production given that the overall proportion of opening is less. However, as the VP mechanism grows, it is plausible that any VP gap would also grow proportionally, ultimately leading to the same impact on speech across age ranges.

One maturational factor that may contribute to differences between the 4- and 8-year-olds is fundamental frequency (f₀). Higher f₀ has been shown to increase listeners’ perception of nasality, even when the degree of nasal coupling is held constant. ³⁸ Because younger children produce speech at substantially higher f₀ than older children, developmental reductions in pitch may partially account for age-related differences in perceived hypernasality. This interpretation is further supported by well-documented decreases in male f₀ following puberty, ³⁹ which may explain the common clinical observation that prepubescent males are perceived as mildly hypernasal but demonstrate perceptually normal resonance after voice change. Importantly, perceptual ratings of hypernasality reflect interactions among nasal coupling, resonance characteristics, and broader acoustic factors,^35,40 underscoring the multifactorial nature of resonance judgments. The fundamental frequency contours for each word had the same temporal structure, but the range and maximum value of the fundamental frequency were appropriate for each age (highest for the 4-year-old audio samples). This is likely to have some effect on results that the maximum VP gap size crossover points occurred at significantly smaller value of area for the 4-year-old stimuli than for the 6- and 8-year-old stimuli.

Limitations and Future Directions

This study has several limitations. First, the CVC stimuli do not capture the complexity of connected speech, and human talkers may use adaptive strategies that are not reflected in the model. Maximum VP gap size alone does not fully determine hypernasality, however, this study represents a step forward in systematically relating degree of nasal coupling to listener perception. It clarifies one component of a multifactorial system and helps delineate which aspects of nasal coupling are most influential for listeners, thereby informing future work on interactive and speech-dynamic factors. Second, VP opening was represented only as cross-sectional area, without reference to the percentage of VP closure, due to limited data on typical port dimensions at rest. This presents a challenge in relating the VP gap sizes reported in this study to studies using nasopharyngoscopy. Third, the model was not configured to simulate nasal air emission, a prominent feature in cleft speech, so the effect of nasal air emission on consonant perception remains unknown; this is a potential enhancement to be added for future studies. Similarly, the potential differences in production and perception of oral [d] versus a nasal substitution for a [d] versus a nasalized [d] are important questions to pursue with a computational modeling approach. Since these questions also represent a significant increase in complexity for designing a systematic listening experiment, we look forward to investigating these questions in the future pending additional knowledge of the potential vocal tract configurational differences. Finally, the listener sample was limited in sociodemographic diversity, which may affect generalizability of nasality perception. Future work should extend these findings using larger and more diverse listener populations, potentially via nationwide crowdsourcing, to examine variability in perceptual judgments. Additional studies could investigate how different degrees of nasal coupling influence word transcription and intelligibility and explore more complex speech contexts to better reflect real-world communication.

Beyond its research applications, computational modeling may offer practical value as an educational tool for training clinicians in perceptual assessment of hypernasality. One of the primary challenges in clinical training is that resonance judgments are inherently multidimensional and influenced by articulation, prosody, vocal intensity, and speaker-specific characteristics. A modeling approach allows systematic manipulation of a single parameter (such as degree of nasal coupling) while holding other variables constant. This controlled exposure may help novice listeners calibrate their internal perceptual scale for hypernasality before confronting the variability of real clinical speech. Additionally, computational stimuli can be generated across different phonetic contexts, speech materials, and prosodic conditions, allowing trainees to practice identifying hypernasality under structured and progressively more complex conditions. Such applications may have the potential to contribute to improved interrater reliability and consistency in clinical decision-making pending future investigations.

Conclusion

Our study demonstrated a shift in perception of an oral stop consonant to a nasal consonant based on nasal coupling areas between 0.014 and 0.035 cm² for simulated vocal tracts of children aged 4, 6 and 8 years. This area is smaller than that observed for syllable contexts as well as for adult talkers. For the 4-year-old vocal tract, significant perceptual differences were particularly notable for consonants in words containing a medial /æ/ vowel. These findings may inform clinical management of VP dysfunction by highlighting the perceptual impact of VP gap size. Future research should explore how listener experience and background influence the perception of increased nasal coupling in the speech of young children.