Can you change a vowel without changing the note?: sources & model
Open real MRI-derived vocal-tract models, swap a vowel while holding the source steady, listen to measured-model sounds, and investigate what growth can change.
The source and model records are available for inspection. No external scientific reviewer has signed off yet.
measured-voice-shapes-1 · content 1 · setup format 1
What supports the explanation?
MRI-derived geometry, measured transfer and reference audio
Scientific Data 7:255. Processed MRI, dental impressions, physical models, measurement method, valid band and Table 3. Article CC BY 4.0; the dataset has its own CC0 license.
CC0 1.0. Three matched speaker-1 configurations: inner surface, printable halves, measured transfer and reference sound. Local source/derivative hashes and geometry transforms are available in the asset record.
Figure 3 pulse C and experiment-1 opening/closing parameters. Our plotted computation is original; no licensed reuse of the published figure is claimed.
Observed vocal-fold length and frequency. The marked change did not coincide with a matching jump in length. Abstract-level evidence; no fabricated participant trajectories or effect sizes.
Computational manipulations of tissue stiffness, medial thickness, initial opening and subglottal pressure. Findings describe that model, not a personal fitted voice.
Eighteen adults performed vocal glides. Source/formant crossover findings bound our independent linear controls. The strenuous research exercises are not the home activity.
Figure 2 contains an actual larynx photo beside the setup illustration. CC BY 4.0; reproduced unmodified. The stacked graphic is not a cycle-frame sequence.
Agency guidance supports avoiding strain and resting when tired or hoarse. The primary paper activity requires no voice production, throat contact or microphone.
Three static vowel shapes from one adult are not scans of developmental change.
The printed shell, 4 mm wall and 10 mm inlet adapter are research objects, not muscle or a life-size glottal opening.
The geometry contains an oral tract and excludes the nasal cavity; it does not model a hum completely.
Opening the model is inspection; all curves refer to the assembled rigid model.
The prescribed flow source is not a tissue-motion, pressure or muscle-control solver.
The plotted data are flow-transfer ratios, not personal loudness or calibrated radiated pressure.
Model sound has a common normalization rule; reference recordings were normalized by the researchers.
The German research vowels are approximate matches to the friendly ah/ee/oo prompts, not exact American-English pronunciation samples.
Measured and finite-element values are different kinds of evidence.
No microphone, personal voice score, identity inference, universal age schedule or health assessment is used.
A static larynx photograph cannot be animated as a measured cycle movie.
Physical-device sound and perception depend on playback conditions; the visual tasks do not require hearing.
Rosenberg pulse C: The prescribed positive flow uses a cosine opening over 0.40 of a cycle, a cosine closing over 0.16, then zero. These parameters come from Rosenberg’s experimental pulse C, not a universal human waveform. Its cycle integral is 0.3018591636 in normalized units.
Complex harmonics retain phase: We compute Fourier coefficients with a 4096-point period and omit DC in audio. Each component at k times the source rate is multiplied by the complex measured transfer. Interpolating real and imaginary values avoids errors at wrapped phase boundaries.
What the transfer means: The dataset characterizes mouth volume flow divided by glottal volume flow for a rigid assembled model. Its experimental method uses pressure measurements and reciprocity. This ratio is not pressure gain at a listener’s ear, and a ratio above one is not a claim that the model creates energy.
A resonance is not a new oscillator: The ah model has a measured resonance near 517 Hz. A 120 Hz periodic source contains 120, 240, 360, 480 Hz and further multiples; filtering changes their complex weights. It does not insert a new 517 Hz sinusoid merely because the curve peaks there.
Formants and resonances: Formants are resonance-related features in a speech spectrum. We label independently measured response peaks as resonance frequencies. Every small local bump in data should not automatically be called a formant.
Only use measured frequency rows: The source files include padded values outside their valid 100–10,000 Hz band. Those values are not evidence of perfect transmission. Runtime data retain valid rows, while the harmonic calculation uses components up to 8,000 Hz.
Actual source files and geometry: The three selected configurations are /aː/, /iː/ and /uː/ from speaker 1. STL coordinates are millimeters. We reindex the original triangles and apply one common transform to each shape’s parts; different vowels are discrete meshes, not invented intermediate anatomy.
Why length alone is insufficient: Harries and colleagues observed gradually increasing vocal-fold length, while the marked frequency change did not coincide with a matching length jump. Zhang’s computational work varies stiffness, thickness, initial opening and pressure. Together they motivate a multi-factor explanation, not a fitted puberty equation.
Growing is not the same as actively stretching: Actively stretching existing tissue can change stiffness as well as dimensions. Tissue growth can also change mass and structure. A string formula with only length varied does not settle either process in layered, airflow-coupled vocal folds.
Source and filter can interact: Titze and colleagues investigated adult vocal exercises near source/formant crossovers and found evidence of interaction. Separate controls are a useful linear approximation here; living phonation can include feedback between tissue and the tract.
What has been checked
Analytical reference cases, conservation or transition invariants, finite drawing commands, bounded setup parsing, discovery and route integrity are checked automatically. These checks do not establish anatomical fidelity, learner outcomes or browser/device compatibility. Independent subject review, learner trials, comprehensive accessibility review and browser video encoding checks remain pending.
Each source supports the associated claim. Sources do not certify this implementation or its visuals.
About the cover illustration
Actual MRI-derived vocal-tract research model, split into its source-defined printable halves. Dresden Vocal Tract Dataset, Birkholz et al. (2020), CC0. Plastic research geometry, not skin or a developmental scan.