Computational Aeroacoustic Prediction of Sibilant Production With Full-Arch Implant Prostheses: A Modeling Methodology and Clinical Validation Protocol Relating Anterior Tooth Geometry, Airflow, and Speech
- Paper type
- Computational modeling methodology with bench and clinical validation protocol
- Status
- Original Work · Working Paper
Contents
Abstract
Background. Speech disturbance, chiefly of sibilants, is common after delivery of maxillary full-arch implant prostheses and usually resolves; in a minority it persists. Clinical studies relate speech outcome to anterior bone volume, prior speech characteristics, and adaptation time, and a systematic review rates the evidence as low quality. Separately, aeroacoustic simulation has reproduced the spectrum of /s/ from individual vocal-tract geometry and has shown that incisor angle changes high-frequency output. These simulations have not been used to predict the speech consequences of a specific prosthesis before it is made.
Objective. To specify a methodology that predicts sibilant acoustics from prosthesis geometry and patient anatomy, and a validation program on bench replicas and in patients.
Framework. The sibilant is modeled as a turbulent jet formed at a tongue–palate constriction, impinging on the incisors, and filtered by the front cavity. Because speakers adjust tongue posture to restore their acoustic target, prediction is posed as a reachability question: given the prosthesis, does a physiologically attainable tongue configuration exist that produces the patient’s pre-treatment sibilant spectrum? Designs for which the target is reachable are predicted to cause transient disturbance at most; designs for which it is not are predicted to cause persistent distortion.
Proposed methods. High-fidelity flow and acoustic simulation on a parametric family of anterior geometries; a reduced-order surrogate; validation against printed replicas in a flow rig; and a prospective cohort in which predicted spectral measures and reachability are compared with recorded speech at delivery and after adaptation.
Central hypothesis. A geometry-based reachability measure predicts persistent sibilant distortion at three months better than geometric descriptors alone.
Conclusions. The physical basis is established and the clinical application is not. The principal uncertainty is the tongue, which no prosthesis design determines. No results are reported.
Keywords: speech; sibilant fricative; implant-supported prosthesis; aeroacoustics; computational fluid dynamics; incisor position; palatal contour; articulation adaptation
1. Introduction
Patients who receive a fixed implant prosthesis in the edentulous maxilla often speak differently the day it is delivered. The sound most affected is /s/. In an early three-year follow-up of 21 patients, initial phonetic problems were common and 94% considered themselves free of speech problems at the end [1]. Later studies using perceptual and instrumental assessment found consonant distortions, especially of sibilants, across types of implant prosthesis [2–4], examined adaptation after immediate loading [5], and described speech problems with fixed provisional restorations in the edentulous maxilla together with a way of managing them [6]. In Mandarin speakers, spectral measures of fricatives and affricates moved toward control values after complete-arch restoration while some distortions remained [7], and improvement was related mainly to preoperative speech, anterior residual bone volume, and adaptation time [8]. A systematic review of speech in implant-supported and removable complete denture wearers graded the evidence with a formal framework and found it limited [9].
The clinical response is empirical. The prosthesis is delivered, the patient is asked to read aloud, and if a lisp or whistle persists after some weeks the palatal contour or incisor position is adjusted by trial. Classical phonetic guides for setting anterior teeth and vertical dimension remain in use [10, 11]. With a provisional prosthesis there is latitude to adjust. With a milled zirconia definitive prosthesis there is very little.
Speech production is a physical process, and for sibilants the physics is reasonably well understood. Air forced through a narrow channel between tongue and palate forms a jet that strikes the teeth, and the turbulence generated there is the sound source; the cavity in front of the constriction shapes its spectrum [12–15]. Over the past decade this understanding has been turned into simulation. Large-eddy simulation in a vocal-tract geometry built from computed tomography and dental casts reproduced the spectrum of an individual’s /s/ and located the source near the incisors [16]. Experiments and simulations in simplified and subject-derived geometries explained differences in sibilant acoustics between individuals [17, 18]. A numerical study varied incisor angle and found that it changed sound amplitude between 8 and 12 kHz by shifting the source position, and its authors noted the relevance to the design of dental prostheses [19].
The two literatures have not met. Clinical studies measure speech after the fact and relate it to coarse descriptors. Simulation studies explain how geometry produces sound in healthy speakers. No study identified has used simulation to predict, from the design file of a full-arch prosthesis and the patient’s anatomy, how that patient’s sibilants will sound, or has tested such a prediction against recordings. The reason is not only technical. A prosthesis fixes the teeth and palate. It does not fix the tongue, and speakers reorganize their articulation to compensate for changes in oral geometry [20, 21]. Any predictive method must deal with that.
This manuscript proposes a methodology that does so by changing what is predicted, and a program for validating it.
2. Background and Current Scientific Evidence
Speech with implant prostheses
The clinical evidence establishes that sibilant distortion is frequent initially, that most patients adapt, and that a minority do not. It identifies associations with prosthesis features in general terms. In the original follow-up, difficulty with /s/ was related to features including the width of the prosthesis in the frontal region [1]. Fixed prostheses without palatal coverage and removable ones with it differ in speech ratings [4]. Speech has been evaluated with mini-implant overdentures [22]. Persistent difficulty has been linked to anatomical limits on where teeth and palatal contour can be placed when anterior bone is deficient [8]. Speakers with a history of articulation disorder adapt less readily to dental prostheses [23]. The systematic review found heterogeneous methods and low-quality evidence [9]. Few studies quantified prosthesis geometry at all, and none in three dimensions against acoustic outcome.
Experimental manipulation of oral geometry
Controlled changes in denture geometry alter sibilants in predictable directions. Tilting maxillary central incisors over a range of 60° in duplicate dentures changed the phonetic pattern of /s/ [24]. A numerical method for speech analysis was applied to dentures with different incisor positions and palatal thicknesses three decades earlier [25]. Artificial palates that thicken the alveolar region degrade fricatives immediately, and speakers compensate partly within minutes and further with targeted practice [20, 21]. These experiments show that anterior tooth position and palatal thickness in the region of the constriction are the sensitive variables, and that compensation is rapid, incomplete, and variable between individuals.
Aeroacoustics of sibilants
The classical account treats frication noise as generated by turbulence in the jet downstream of a constriction, with source strength depending on flow velocity and on the presence of an obstacle in the jet’s path [12, 13]. For /s/, the lower and upper incisors act as the obstacle. Theoretical treatments based on aerodynamic sound theory [26, 27] have described the source mechanisms and their scaling [14, 15]. Simulation has followed: large-eddy simulation coupled with an acoustic analogy in realistic and simplified tract geometries [16–18]; parametric variation of incisor angle [19]; and patient-specific studies of abnormal sibilant production with velopharyngeal openings [28]. Physical articulation simulators with replaceable anatomy have been used to study speech after maxillectomy [29], and imaging methods that include the teeth in vocal-tract models are improving [30].
Acoustic description of sibilants
Sibilant spectra are summarized by the frequency of the main spectral peak and by spectral moments: center of gravity, dispersion, skewness, and kurtosis [31, 32]. Refinements separate energy in low, mid, and high bands and identify a mid-frequency peak that is more robust for /s/ [33]. These measures are computed with standard software [34] and are the endpoints used in the clinical studies cited [7]. They provide the quantities a predictive model must reproduce.
Summary
It is established that sibilants are sensitive to anterior tooth position and palatal contour, that the mechanism is turbulent noise generation at the incisors filtered by the front cavity, and that this can be simulated from geometry. It is established clinically that most patients adapt and some do not. No method predicts for an individual prosthesis which outcome to expect.
3. Research Gap
Three gaps separate the available simulation capability from clinical use. First, simulations have been run on the geometry of healthy speakers producing a sibilant, with the tongue shape taken from imaging during phonation. For a patient about to receive a prosthesis, the tongue shape with the new prosthesis is not known, because it has not happened yet. Second, high-fidelity aeroacoustic simulation takes hours to days per case, which precludes its use in an iterative design process. Third, no validation exists that compares predicted with observed speech for prostheses, at the bench or in patients.
A conceptual gap underlies the first. Predicting the acoustic output of a fixed geometry is a forward problem. Predicting what a speaker will do with a new geometry is a problem of motor adaptation. The former has been solved for selected cases. The latter has not been formulated for prosthesis design.
4. Research Objective and Central Hypothesis
The objective is to develop and validate a method that, from the digital design of a maxillary full-arch prosthesis and the patient’s pre-treatment anatomy and speech, predicts whether the patient will be able to produce an acoustically normal sibilant with the prosthesis.
The central hypothesis is that a reachability measure, defined as the smallest achievable distance between the simulated sibilant spectrum and the patient’s pre-treatment target spectrum over a set of physiologically attainable tongue configurations, predicts persistent sibilant distortion three months after delivery better than a model based on geometric descriptors of the prosthesis alone.
Secondary hypotheses: (a) at the bench, simulated spectral peak frequency and center of gravity agree with measurements on printed replicas within limits comparable to published agreement for natural geometries; (b) the acoustic distance between the patient’s sibilant at delivery and their target is predicted by the simulated output for their pre-treatment tongue configuration placed in the new geometry, that is, before adaptation; (c) the reduced-order surrogate reproduces the high-fidelity predictions of the spectral measures within a prespecified tolerance.
The central hypothesis is refuted if reachability adds nothing to geometric descriptors, or if persistent distortion occurs as often in designs predicted reachable as in those predicted unreachable.
5. Scientific and Theoretical Framework
Source–filter description
The far-field sound pressure spectrum of a sibilant is written as the product of a source spectrum and a transfer function:
P(f)=S(f; U_j, d_c, ℓ_o, θ_i)⋅T(f; G_f)⋅R(f) (1)
where S is the spectrum of the aeroacoustic source, depending on jet velocity U_j at the constriction, constriction hydraulic diameter d_c, distance from constriction exit to the incisal obstacle ℓ_o, and incisor inclination θ_i; T is the transfer function of the front cavity with geometry G_f (the space bounded by the tongue tip, the lingual surfaces of the anterior teeth, the incisal gap, and the lips); and R is the radiation characteristic at the lips. The prosthesis determines the lingual surfaces of the teeth, the incisal positions, and the anterior palatal contour. The speaker determines the constriction, the tongue-tip position, the jaw opening, and the lip posture.
Flow regime and scaling
For an intraoral pressure p_o during a fricative, the jet velocity is approximately
U_j≈√((2 p_o)/ρ) (2)
with ρ the air density. For pressures of several hundred pascals, U_j is of the order of 30 to 40 m/s. With a constriction a few millimeters across, the Reynolds number U_jd_c/ν is of the order of several thousand to ten thousand, and the jet is turbulent; volume flow rates of several hundred cubic centimeters per second, as used in published simulations [16], are consistent with this. The Mach number is about 0.1. In this regime, sound generated by unsteady forces on an obstacle in the jet scales approximately with the sixth power of velocity, and sound from free turbulence with the eighth [14, 26]. The obstacle source dominates for /s/, which is why incisor position and inclination matter [13, 19].
The transfer function contributes spectral shape. The front cavity behaves to a first approximation as a short tube closed at the constriction and open at the lips, with a lowest resonance near
f_1≈c/(4 L_f) (3)
for effective length L_f and sound speed c. A front cavity of 1.5 to 2 cm places f_1 at roughly 4 to 6 kHz, the region of the main spectral peak of /s/. Lengthening the front cavity by a few millimeters, by moving the incisors labially or the constriction posteriorly, lowers the peak by several hundred hertz. These are order-of-magnitude statements from equations (2) and (3) with standard values for air; they are consistent with published simulations and are not themselves results.
Design-controlled and speaker-controlled variables
Table 1 separates the two sets of variables. The separation is the core of the framework: the design fixes the first set, and the speaker searches over the second.
| Variable | Set by | Acoustic role |
|---|---|---|
| Anteroposterior position and inclination of maxillary incisors | Design | Obstacle distance and angle; front-cavity length [19, 24] |
| Incisal edge level; overjet and overbite | Design | Incisal gap forming the exit; obstacle exposure |
| Palatal contour thickness in the alveolar zone | Design | Space in which the tongue forms the groove [20] |
| Lingual form of anterior teeth | Design | Front-cavity shape |
| Gaps beneath or between prosthesis units | Design | Parallel leak paths; additional sources |
| Occlusal vertical dimension | Design | Jaw opening range for the closest speaking space [11] |
| Tongue-tip position and groove shape | Speaker | Constriction area and location |
| Jaw opening and mandibular incisor position during /s/ | Speaker, within limits set by design | Incisal gap; obstacle geometry |
| Lip posture | Speaker | Radiation; front-cavity length |
| Driving pressure | Speaker | Jet velocity; overall level |
Prediction as reachability
Let g denote the design-controlled geometry and q the speaker-controlled configuration, constrained to a set Q(g) of physiologically attainable configurations given the design: the tongue cannot pass through the palate, the groove cannot be narrower than tissue permits, and the mandibular incisors must clear the maxillary ones. Let m(g,q) be the vector of acoustic measures (spectral peak, center of gravity, skewness, level in defined bands) produced by the model, and m^* the patient’s target, taken from pre-treatment recordings if the patient’s speech was acceptable then, or from normative values for age, sex, and language otherwise. The reachability distance is
D(g)=minq∈Q(g) (W(m(g,q)−m^*)) (4)
with W a weighting that expresses each measure in units of its just-noticeable or normative variation. A design with small D permits a normal sibilant; the patient may need time to find the configuration. A design with large D does not permit one whatever the patient does. A second quantity, the adaptation distance, is the change in configuration required:
A(g)=(q_(opt)(g)−q_0) (5)
where q_0 is the patient’s pre-treatment configuration and q_(opt) the minimizer in equation (4). The framework predicts that persistent distortion relates to D, and the duration of the transient to A.
6. Methodology
Stage 1: high-fidelity simulation on a parametric family
Geometry. A parametric model of the anterior oral cavity from the constriction to the lips, with parameters for incisor anteroposterior position, inclination, and vertical position; palatal thickness in the alveolar zone; overjet and overbite; constriction area, location, and groove shape; jaw opening; and lip aperture. Baseline dimensions are taken from published subject-derived geometries [16, 18].
Simulation. Compressible large-eddy simulation, or incompressible large-eddy simulation with an acoustic analogy [27], at flow rates spanning the physiological range for /s/. Mesh and time-step refinement studies establish that the spectral measures of interest are resolved to the frequency required, at least 12 kHz given the reported sensitivity of the 8 to 12 kHz band to incisor angle [19]; discretization uncertainty is reported [35]. A Latin hypercube design over the parameter space provides the training set, with size determined by the convergence of the surrogate.
Verification against published cases. The simplified-tract configuration for which experimental and numerical spectra have been published [17] is reproduced first, as a check on the solver and settings.
Stage 2: reduced-order surrogate
A statistical emulator (Gaussian process or equivalent) is trained to map geometric and flow parameters to the acoustic measures, with predictive uncertainty. In parallel, a physics-based reduced model implements equation (1) with a parametrized source and a one-dimensional or low-order acoustic model of the front cavity, calibrated to the simulations. The two are compared on held-out simulations; the surrogate used for clinical prediction is the one with smaller error, and its uncertainty is carried into the reachability computation. Model credibility is documented following ASME V&V 40 [36].
Stage 3: bench validation
Replicas. Twelve or more anterior-cavity geometries spanning the parameter space, including pairs that differ only in incisor inclination and only in palatal thickness, are printed at full scale with smooth internal surfaces. A fixed tongue insert defines the constriction.
Rig. Regulated airflow through a settling chamber into the replica in an anechoic or semi-anechoic enclosure, with a calibrated microphone at a fixed far-field position and upstream pressure and flow rate recorded, following published experimental arrangements [17, 18].
Comparison. Measured and simulated spectral peak frequency, center of gravity, and band levels are compared by Bland–Altman limits of agreement [37]. Secondary hypothesis (a) is assessed here. Systematic bias is used to recalibrate the surrogate. Random disagreement sets a floor on the precision of any clinical prediction.
Stage 4: prospective clinical validation
Design. Prospective cohort, reported to STROBE [38].
Participants. Adult native speakers of the study language, without hearing impairment beyond a screening threshold and without pre-existing speech or neurological disorder, receiving a maxillary full-arch fixed implant prosthesis with an interim stage. A target of 60 participants is proposed, on the reasoning in Section 6.6.
Pre-treatment data. Intraoral scans of both arches and occlusion with the existing dentition or prosthesis; audio recordings of a standard word list and passage containing sibilants in varied vowel contexts, made with a head-mounted microphone at fixed distance in a quiet room at a sampling rate sufficient for analysis to 16 kHz; and, in a subsample, tongue configuration during sustained /s/ by ultrasound or by electropalatography on a duplicate of the existing prosthesis. Where no imaging of the tongue is obtained, q_0 is estimated by inverting the model on the pre-treatment recording.
Prediction. Before delivery, the design file of the interim prosthesis is combined with patient anatomy to compute D, A, and the predicted spectrum for the unadapted configuration. Predictions are time-stamped and locked.
Outcome assessment. Recordings at delivery, two weeks, and three months, and after delivery of the definitive prosthesis. Acoustic measures by a standard scripted analysis [32–34]. Perceptual rating of sibilant accuracy by two speech-language pathologists blinded to time point and prediction, with inter-rater reliability reported [39]. Patient-reported speech difficulty.
Primary outcome. Persistent sibilant distortion at three months, defined by perceptual rating, with the acoustic distance from target as a continuous co-primary measure.
Clinical adjustments. Any modification of the prosthesis for speech during follow-up is recorded with its geometry, and the prediction is recomputed for the modified design; such cases provide within-patient tests of the model.
Statistical analysis
Central hypothesis. Logistic regression of persistent distortion on D, compared with a model containing geometric descriptors only (incisor position change, palatal thickness change, vertical dimension change), by discrimination and calibration with bootstrap optimism correction [40]. With an expected minority of events, the analysis is exploratory in the statistical sense, and the continuous co-primary outcome carries the main inferential weight: the association between D and acoustic distance from target at three months.
Secondary hypothesis (b). Agreement between predicted unadapted spectral measures and those recorded at delivery, by Bland–Altman analysis.
Adaptation. The time course of acoustic distance from target is modeled with an exponential decay, and its time constant is related to A.
Sample size
For the continuous co-primary analysis, detecting a correlation of 0.35 between D and acoustic distance at three months with 80% power at two-sided α of 0.05 requires about 60 participants. The value 0.35 is an assumption about the smallest association that would make the method useful. If persistent distortion occurs in 10% to 20% of participants, a range chosen for planning and not taken from a specific source, 6 to 12 events are expected, too few for stable estimation of discrimination; the binary analysis is reported with that limitation.
Bias and reproducibility
Predictions are locked before outcome data exist. Raters are blinded. Recording conditions are standardized across sessions. Clinicians are not shown predictions, to avoid influencing adjustments. Parametric geometries, solver settings, surrogate code, replica files, and de-identified acoustic measures are to be released.
Ethics
Institutional review board approval and consent, including consent for storage of voice recordings, which are identifiable. Imaging of the tongue uses non-ionizing methods.
Language and stimulus material
Sibilant inventories and their acoustic targets differ between languages, and the clinical studies cited were conducted in Swedish, Dutch, and Mandarin among others. The cohort is recruited in a single language, with stimuli chosen to place /s/ and a contrasting sibilant in word-initial, medial, and final positions adjacent to high, low, rounded, and unrounded vowels, because lip rounding lengthens the front cavity and lowers the spectral peak. Sustained sibilants are recorded as well, since they correspond most closely to the steady-state condition that the model represents. Targets and normative ranges are specific to the language and must be re-established before the method is applied in another.
7. Analytical Development
Sensitivity of the spectral peak to incisor position
From equation (3), a small change ΔL_f in effective front-cavity length shifts the lowest resonance by
Δf_1/f_1≈−ΔL_f/L_f (6)
For L_f of 18 mm, moving the incisal edges 2 mm labially lowers f_1 by about 11%, or roughly 500 Hz at 4.8 kHz. A shift of that size is comparable to the differences in center of gravity reported between pre- and post-restoration speech in clinical studies, without implying that it is their cause. The speaker can offset the shift by advancing the constriction by the same amount, if the palatal contour allows the tongue tip to move forward. If the anterior palatal region of the prosthesis is thick, the constriction cannot advance, and the shift cannot be compensated. Equation (6) therefore implies that labial positioning of incisors and thickening of the anterior palate, each tolerable alone, are jointly unfavorable. The combination arises when implants emerge palatally in a resorbed maxilla and teeth must be set labially for lip support, the anatomical situation that the clinical literature associates with persistent difficulty [8].
Sensitivity of level to constriction and obstacle
For an obstacle source scaling as U_j^6, a change in jet velocity of 10% changes sound power by about 77%, or 2.5 dB. At fixed driving pressure, equation (2) gives a jet velocity independent of constriction area, so level is governed by how much of the jet strikes the obstacle and at what distance. Moving the obstacle away from the constriction exit allows the jet to spread and decelerate before impact. For a turbulent jet whose centerline velocity decays inversely with distance beyond a potential core of a few constriction diameters, impact velocity falls roughly as d_c/ℓ_o once ℓ_o exceeds that core length, and source power as the sixth power of that ratio. An increase in obstacle distance from four to six constriction diameters would reduce source power by an order of magnitude under this idealization. The actual sensitivity is lower, because the jet is confined and the geometry is not a free jet, and this is what the simulations are for. The direction agrees with the finding that incisor angle alters high-frequency amplitude by moving the effective source location [19].
Leak paths
A gap beneath a hybrid prosthesis or between units provides a parallel path for air. If its acoustic and aerodynamic resistance is comparable to that of the constriction, a fraction of the flow bypasses the sibilant channel, lowering jet momentum and adding a second, uncontrolled source with its own spectrum, heard as a whistle or as slurring. For two parallel orifices under the same pressure, flow divides approximately in proportion to their areas. A leak of 5 mm² beside a constriction of 10 mm² would carry about a third of the flow. No tongue configuration removes a leak the tongue cannot reach, so leak paths enter the reachability computation as part of g and are expected to be among the strongest determinants of D. This prediction is easily tested at the bench by opening and closing a controlled gap in a replica.
What reachability can and cannot predict
Reachability addresses whether a normal sibilant is physically possible. It does not predict whether a given patient will find the required configuration, how long that will take, or whether a listener will judge the result acceptable. Patients differ in sensorimotor adaptability [21, 23], and that difference is outside the model. The framework therefore makes an asymmetric claim: large D should reliably predict persistent distortion, while small D predicts only that adaptation is possible. In terms of test performance, the method is expected to have higher specificity for good outcome than sensitivity, and the analysis reports both.
8. Discussion
The proposal connects a mature physical understanding of sibilant production to a clinical problem that is still managed by trial. Its distinguishing feature is the treatment of the speaker. Earlier simulation work fixed the tongue and asked what sound results. For prosthesis design, the tongue is the unknown, and the useful question is whether any attainable tongue posture gives an acceptable sound.
The approach has real weaknesses. The set of attainable configurations is defined by constraints that are only partly known. The tongue is a deformable body with its own mechanics, and which groove shapes it can form against a given palatal contour is not something a geometric constraint captures well. If the attainable set is drawn too generously, every design appears reachable and the method has no discriminating power. If too tightly, the method predicts failure where patients succeed. The clinical validation is the only way to calibrate it, and the within-patient cases in which a prosthesis is adjusted are especially valuable for that purpose.
The target is also uncertain. Many patients arrive with a failing dentition or an ill-fitting denture and speech that is already altered, so their pre-treatment sibilant may be a poor target. Normative targets depend on language, dialect, sex, and age. The acoustic measures that best track perceived quality are debated [33]. The weighting in equation (4) embeds assumptions about perceptual relevance that should be tested against the listeners’ ratings.
Computation is a practical obstacle. Large-eddy simulation resolving sound to 12 kHz in a geometry with millimeter-scale features is expensive, and the surrogate must cover a parameter space of ten or more dimensions. The reduced physical model is the fallback. Whether either surrogate attains the accuracy needed to distinguish designs that differ by a millimeter is an open question, and the bench stage is designed to answer it before patients are involved.
Alternative approaches deserve consideration. A purely empirical model trained on geometric descriptors and outcomes from many patients would avoid simulation entirely. It needs a data set that does not exist, though one clinical group has proposed data collection toward such prediction models [8]. A physical articulation simulator with interchangeable prosthesis replicas [29] could serve as a chairside or laboratory test, trading computation for hardware. The simulation approach has the advantage of working from the design file alone, before anything is fabricated, and of explaining its predictions in terms a clinician can act on: front-cavity length, obstacle distance, tongue space, leak.
It is fair to ask whether the problem warrants the effort, since most patients adapt. The case rests on the minority who do not, on the cost of remaking a definitive full-arch prosthesis, and on the trend toward fully digital workflows in which the interim prosthesis that once allowed phonetic trial is abbreviated or omitted.
9. Clinical Significance
If validated, the method would give a designer a speech-related check before fabrication: a flag when a combination of incisor position, palatal thickness, and vertical dimension places the patient’s sibilant target out of reach, and an indication of which change brings it back. The analytical results already point to a specific combination to avoid, labial incisor position together with a thick anterior palatal contour, and to leak paths as a dominant factor. For patients whose design is reachable with a large required change in articulation, the prediction of a longer adaptation could inform counseling and referral for speech therapy.
For patients with anatomical constraints that make an unfavorable combination unavoidable, the method would at least quantify the compromise, which bears on the choice between fixed and removable designs.
These uses depend on validation that has not been done.
10. Limitations
The attainable configuration set is poorly constrained. The pre-treatment target may be unrepresentative. Only sibilants are modeled; other sounds affected by prostheses, such as alveolar stops and labiodentals, are not. The model treats steady frication and ignores coarticulation and dynamics. High-fidelity simulation is costly, and surrogate accuracy is unproven. Bench replicas have rigid walls and a fixed tongue. The clinical cohort is small, and persistent distortion will be infrequent. Results are language-specific. Tongue imaging is limited to a subsample. Scaling arguments in Section 7 use free-jet and simple-resonator idealizations and standard physical constants; they indicate direction and order of magnitude only. The literature was assessed from abstracts and index records.
11. Future Validation
A larger multicenter cohort would be needed to estimate predictive performance for the binary outcome. A within-patient experiment using two interim prostheses that differ in a single design variable would test causal predictions directly. Extension to other languages requires language-specific targets. Coupling the acoustic model to a biomechanical tongue model would replace the geometric constraint set with a physical one. Independent reproduction of the bench comparison with replicas printed from the shared files would establish that the simulation settings are transferable.
12. Conclusion
Sibilant distortion after maxillary full-arch rehabilitation has a physical explanation, and the tools to simulate that physics exist. What has been missing is a way to use them for a patient whose tongue has not yet met the prosthesis. Posing prediction as a question of reachability, whether any attainable articulation reproduces the patient’s sibilant with the proposed design, accommodates adaptation and yields quantities a designer can use. Simple scaling shows why labially placed incisors combined with anterior palatal bulk, and leak paths the tongue cannot seal, should be the configurations most likely to defeat adaptation. The simulations, bench tests, and clinical cohort described would determine whether the method predicts what patients experience. None has been carried out.
Cite this paper
Jain S, Gupta A. Computational Aeroacoustic Prediction of Sibilant Production With Full-Arch Implant Prostheses: A Modeling Methodology and Clinical Validation Protocol Relating Anterior Tooth Geometry, Airflow, and Speech. Working paper. Center for Implant Dentistry; 2026. Available from: https://bayareaimplantdentistry.com/clinician-research-and-publications/computational-acoustics-full-arch-speech
References
- Lundqvist S, Haraldson T, Lindblad P. Speech in connection with maxillary fixed prostheses on osseointegrated implants: a three-year follow-up study. Clin Oral Implants Res. 1992;3(4):176-80. doi:10.1034/j.1600-0501.1992.030404.x (opens in a new tab). PMID: 1298432 (opens in a new tab).
- VAN Lierde K, Browaeys H, Corthals P, Mussche P, VAN Kerkhoven E, DE Bruyn H. Comparison of speech intelligibility, articulation and oromyofunctional behaviour in subjects with single-tooth implants, fixed implant prosthetics or conventional removable prostheses. J Oral Rehabil. 2012;39(4):285-93. doi:10.1111/j.1365-2842.2011.02282.x (opens in a new tab). PMID: 22239248 (opens in a new tab).
- Jacobs R, Manders E, Van Looy C, Lembrechts D, Naert I, van Steenberghe D. Evaluation of speech in patients rehabilitated with various oral implant-supported prostheses. Clin Oral Implants Res. 2001;12(2):167-73. doi:10.1034/j.1600-0501.2001.012002167.x (opens in a new tab). PMID: 11251667 (opens in a new tab).
- Heydecke G, McFarland DH, Feine JS, Lund JP. Speech with maxillary implant prostheses: ratings of articulation. J Dent Res. 2004;83(3):236-40. doi:10.1177/154405910408300310 (opens in a new tab). PMID: 14981126 (opens in a new tab).
- Molly L, Nackaerts O, Vandewiele K, Manders E, van Steenberghe D, Jacobs R. Speech adaptation after treatment of full edentulism through immediate-loaded implant protocols. Clin Oral Implants Res. 2008;19(1):86-90. doi:10.1111/j.1600-0501.2007.01305.x (opens in a new tab). PMID: 17944967 (opens in a new tab).
- Collaert B, Van Dessel J, Konings M, Nackaerts O, Zink I, Slagmolen P, et al. On Speech Problems with Fixed Restorations on Implants in the Edentulous Maxilla: Introduction of a Novel Management Concept. Clin Implant Dent Relat Res. 2015;17 Suppl 2:e745-50. doi:10.1111/cid.12309 (opens in a new tab). PMID: 25873135 (opens in a new tab).
- Zhang Y, Liu J, Yu D, Ding H, Wu Y. Articulation distortion in Mandarin-speaking individuals with complete arch maxillary implant-supported fixed dental prostheses. J Prosthet Dent. 2025;134(3):704-711. doi:10.1016/j.prosdent.2023.10.008 (opens in a new tab). PMID: 37978009 (opens in a new tab).
- Zhang Y, Ding H, Gao H, Liu J, Wang F, Wu Y. Factors influencing speech improvement following maxillary complete-arch implant-supported restorations: A retrospective study. Clin Oral Implants Res. 2024;35(10):1240-1250. doi:10.1111/clr.14311 (opens in a new tab). PMID: 38867397 (opens in a new tab).
- Meira IA, Gama LT, Prado-Tozzi DA, Pinheiro MA, Rodrigues Garcia RCM. Speech in implant-supported and removable complete denture wearers: A systematic review. J Prosthet Dent. 2022;128(6):1230-1238. doi:10.1016/j.prosdent.2021.03.006 (opens in a new tab). PMID: 33865562 (opens in a new tab).
- Pound E. Let /S/ be your guide. J Prosthet Dent. 1977;38(5):482-9. doi:10.1016/0022-3913(77)90022-1 (opens in a new tab). PMID: 269949 (opens in a new tab).
- Silverman MM. The speaking method in measuring vertical dimension. 1952. J Prosthet Dent. 2001;85(5):427-31. doi:10.1067/mpr.2001.116139 (opens in a new tab). PMID: 11357066 (opens in a new tab).
- Stevens KN. Airflow and Turbulence Noise for Fricative and Stop Consonants: Static Considerations. The Journal of the Acoustical Society of America. 1971;50(4B):1180-1192. doi:10.1121/1.1912751 (opens in a new tab).
- Shadle CH. The effect of geometry on source mechanisms of fricative consonants. Journal of Phonetics. 1991;19(3-4):409-424. doi:10.1016/s0095-4470(19)30332-8 (opens in a new tab).
- Howe MS, McGowan RS. Aeroacoustics of [s]. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences. 2005;461(2056):1005-1028. doi:10.1098/rspa.2004.1405 (opens in a new tab).
- Krane MH. Aeroacoustic production of low-frequency unvoiced speech sounds. J Acoust Soc Am. 2005;118(1):410-27. doi:10.1121/1.1862251 (opens in a new tab). PMID: 16119362 (opens in a new tab).
- Nozaki K, Yoshinaga T, Wada S. Sibilant /s/ simulator based on computed tomography images and dental casts. J Dent Res. 2014;93(2):207-11. doi:10.1177/0022034513514586 (opens in a new tab). PMID: 24300309 (opens in a new tab).
- Yoshinaga T, Nozaki K, Wada S. Experimental and numerical investigation of the sound generation mechanisms of sibilant fricatives using a simplified vocal tract model. Physics of Fluids. 2018;30(3). doi:10.1063/1.5013632 (opens in a new tab).
- Yoshinaga T, Nozaki K, Wada S. Aeroacoustic analysis on individual characteristics in sibilant fricative production. J Acoust Soc Am. 2019;146(2):1239. doi:10.1121/1.5122793 (opens in a new tab). PMID: 31472528 (opens in a new tab).
- Lu H, Yoshinaga T, Li C, Nozaki K, Iida A, Tsubokura M. Numerical investigation of effects of incisor angle on production of sibilant /s/. Sci Rep. 2021;11(1):16720. doi:10.1038/s41598-021-96173-2 (opens in a new tab). PMID: 34408209 (opens in a new tab).
- McFarland DH, Baum SR, Chabot C. Speech compensation to structural modifications of the oral cavity. J Acoust Soc Am. 1996;100(2 Pt 1):1093-104. doi:10.1121/1.416286 (opens in a new tab). PMID: 8759962 (opens in a new tab).
- Baum SR, McFarland DH. The development of speech adaptation to an artificial palate. J Acoust Soc Am. 1997;102(4):2353-9. doi:10.1121/1.419619 (opens in a new tab). PMID: 9348694 (opens in a new tab).
- Fonteyne E, Van Doorne L, Becue L, Matthys C, Bronckhorst E, De Bruyn H. Speech evaluation during maxillary mini-dental implant overdenture treatment: A prospective study. J Oral Rehabil. 2019;46(12):1151-1160. doi:10.1111/joor.12852 (opens in a new tab). PMID: 31271663 (opens in a new tab).
- Hamlet SL, Stone M. Speech adaptation to dental prostheses: the former lisper. J Prosthet Dent. 1982;47(5):564-9. doi:10.1016/0022-3913(82)90311-0 (opens in a new tab). PMID: 7045340 (opens in a new tab).
- Runte C, Lawerino M, Dirksen D, Bollmann F, Lamprecht-Dinnesen A, Seifert E. The influence of maxillary central incisor position in complete dentures on /s/ sound production. J Prosthet Dent. 2001;85(5):485-95. doi:10.1067/mpr.2001.114448 (opens in a new tab). PMID: 11357076 (opens in a new tab).
- Petrović A. Speech sound distortions caused by changes in complete denture morphology. J Oral Rehabil. 1985;12(1):69-79. doi:10.1111/j.1365-2842.1985.tb00622.x (opens in a new tab). PMID: 3886862 (opens in a new tab).
- Lighthill MJ. On sound generated aerodynamically I. General theory. Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences. 1952;211(1107):564-587. doi:10.1098/rspa.1952.0060 (opens in a new tab).
- Ffowcs Williams JE, Hawkings DL. Sound generation by turbulence and surfaces in arbitrary motion. Philosophical Transactions of the Royal Society of London. Series A, Mathematical and Physical Sciences. 1969;264(1151):321-342. doi:10.1098/rsta.1969.0031 (opens in a new tab).
- Sundström E, Oren L. Change in aeroacoustic sound mechanism during sibilant sound with different velopharyngeal opening sizes. Med Biol Eng Comput. 2021;59(4):937-945. doi:10.1007/s11517-021-02356-4 (opens in a new tab). PMID: 33797695 (opens in a new tab).
- Elbashti ME, Hattori M, Sumita YI, Taniguchi H. Evaluation of articulation simulation system using artificial maxillectomy models. J Oral Rehabil. 2015;42(9):678-84. doi:10.1111/joor.12306 (opens in a new tab). PMID: 25975670 (opens in a new tab).
- Ramtilak S, Lu A, Holmes J, Atha J, Hoffman EA, Story B, et al. Incorporating Dental Anatomy Into MRI-Based Vocal Tract Models Using Zero Echo Time Imaging. J Voice. 2026. doi:10.1016/j.jvoice.2026.04.035 (opens in a new tab). PMID: 42140867 (opens in a new tab).
- Forrest K, Weismer G, Milenkovic P, Dougall RN. Statistical analysis of word-initial voiceless obstruents: preliminary data. J Acoust Soc Am. 1988;84(1):115-23. doi:10.1121/1.396977 (opens in a new tab). PMID: 3411039 (opens in a new tab).
- Jongman A, Wayland R, Wong S. Acoustic characteristics of English fricatives. J Acoust Soc Am. 2000;108(3 Pt 1):1252-63. doi:10.1121/1.1288413 (opens in a new tab). PMID: 11008825 (opens in a new tab).
- Koenig LL, Shadle CH, Preston JL, Mooshammer CR. Toward improved spectral measures of /s/: results from adolescents. J Speech Lang Hear Res. 2013;56(4):1175-89. doi:10.1044/1092-4388(2012/12-0038) (opens in a new tab). PMID: 23785194 (opens in a new tab).
- Boersma P, Weenink D. Praat: doing phonetics by computer [computer program]. Amsterdam: University of Amsterdam.
- Roache PJ. Perspective: A Method for Uniform Reporting of Grid Refinement Studies. Journal of Fluids Engineering. 1994;116(3):405-413. doi:10.1115/1.2910291 (opens in a new tab).
- American Society of Mechanical Engineers. Assessing credibility of computational modeling through verification and validation: application to medical devices. ASME V&V 40-2018. New York: ASME; 2018.
- Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;1(8476):307-10. PMID: 2868172 (opens in a new tab).
- von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet. 2007;370(9596):1453-7. doi:10.1016/S0140-6736(07)61602-X (opens in a new tab). PMID: 18064739 (opens in a new tab).
- Koo TK, Li MY. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med. 2016;15(2):155-63. doi:10.1016/j.jcm.2016.02.012 (opens in a new tab). PMID: 27330520 (opens in a new tab).
- Van Calster B, McLernon DJ, van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. BMC Med. 2019;17(1):230. doi:10.1186/s12916-019-1466-7 (opens in a new tab). PMID: 31842878 (opens in a new tab).

