💬 NLP

Articulatory strategy as a source of variation in acoustic vowel dynamics

This study utilizes ultrasound tongue imaging of 36 Northern-Anglo English speakers to demonstrate that distinct individual articulatory strategies for producing the vowel /i/ systematically predict the shape and timing of formant dynamics in diphthongs, thereby linking specific tongue movements to acoustic variation and speaker individuality.

Original authors: Patrycja Strycharczuk, Justin J. H. Lo, Sam Kirkham

Published 2026-05-25

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Patrycja Strycharczuk, Justin J. H. Lo, Sam Kirkham

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Idea: Why Your Voice Sounds Like You

Imagine you and a friend are both trying to hit a specific note on a piano. You both hit the exact same key (the target sound), but the way your fingers move to get there is different. Maybe your friend uses a quick, sharp tap, while you use a slow, gliding press. Even though the note sounds the same at the end, the journey to get there sounds different.

This paper investigates that "journey" in human speech. Specifically, it looks at vowel dynamics—how our voices change as we move from one sound to another (like in a diphthong, which is a sliding vowel sound like the "oy" in "boy").

The researchers wanted to know: Does the unique way a person moves their tongue (their "articulatory strategy") leave a fingerprint on the sound of their voice?

The Setup: The "Tongue Map"

The study looked at 36 people from Northern England. The researchers used ultrasound technology (like a medical sonogram) to take moving pictures of the inside of their mouths while they spoke. This allowed them to see exactly how each person shaped their tongue.

They focused on the sound /i/ (like the "ee" in "see"). They found that even though everyone was trying to say the same sound, their tongues looked different:

Some people had tongues that were very domed (curved high up like a hill).
Some had tongues that were flatter.
Some pushed their tongues forward, while others kept them back.

Think of this like different people driving the same car. One driver might sit very close to the steering wheel, while another sits back. They are both driving the same car, but their "driving style" (or in this case, tongue shape) is unique to them.

The Experiment: The Sliding Sounds

The researchers then listened to these same people say sliding vowel sounds (diphthongs) like "bade" (bay-uhd), "bide" (bye-uhd), and "buoyed" (boy-uhd). These sounds start in one place and slide toward the "ee" sound.

They asked: If a person has a very domed tongue for the "ee" sound, does that change how their voice slides into that sound?

The Findings: The "Speed Limit" of the Mouth

The answer was yes. The study found a clear link between the shape of the tongue and the speed of the sound.

Here is the analogy: Imagine two runners trying to reach the same finish line in exactly the same amount of time.

Runner A has to run a short distance.
Runner B has to run a much longer distance.

To finish at the same time, Runner B has to run faster and accelerate more quickly than Runner A.

The paper found that people with certain tongue shapes (like a very domed tongue) had to move their tongue a longer distance to get to the target sound. Because they had to cover more ground in the same amount of time, their tongues had to move faster.

This faster movement created a specific sound signature:

Earlier transitions: The sound started changing sooner.
Steeper changes: The pitch of the voice changed more sharply and quickly.

Why This Matters

For a long time, scientists thought that if two people said the same word, the sound would be almost identical because our brains are good at "compensating" for our different mouth shapes. We all adjust to make sure the sound comes out right.

This paper shows that while we do adjust to make the target sound correct, we can't hide our movement style. Just like you can't hide how you run just because you're running to the same finish line, you can't hide how you move your tongue just because you're making the same vowel.

The Conclusion

The study concludes that the unique "fingerprint" in a person's voice isn't just about the static sounds they make; it's about how they get there.

The Strategy: Your unique anatomy (the size and shape of your mouth) forces you to use a specific movement strategy.
The Result: That strategy changes the speed and shape of the sound wave as it travels.
The Takeaway: Even when we try to sound the same, our bodies tell a different story through the speed and dynamics of our speech.

The researchers did not test this for forensic identification (like solving crimes) or medical diagnosis in this specific paper; they simply established the scientific link between how the tongue moves and how the sound changes. They proved that movement creates individuality.

Technical Summary: Articulatory Strategy as a Source of Variation in Acoustic Vowel Dynamics

Problem Statement
Acoustic vowel dynamics contain speaker-identifying information, a fact well-documented in forensic phonetics where dynamic formant transitions often outperform static measurements in speaker recognition. The prevailing hypothesis attributes this individuality to distinct articulatory strategies—specifically, the unique movement patterns speakers employ to reach acoustic targets. However, direct empirical evidence linking specific articulatory strategies to systematic variations in acoustic formant dynamics has been scarce. While existing literature extensively documents articulatory compensation (where speakers adjust their articulators to maintain acoustic invariance despite anatomical differences), it remains unclear how these compensatory strategies influence the dynamic properties of the acoustic signal, particularly in inherently dynamic sounds like diphthongs. This study addresses the gap by investigating whether the articulatory strategy used for a static palatal vowel (/i/) systematically conditions the formant dynamics of I-diphthongs.

Methodology
The study utilizes a secondary analysis of the TarDiS corpus, comprising ultrasound tongue imaging (UTI) and acoustic data from 36 speakers of Northern-Anglo English (after excluding four speakers with monophthongal realizations of the /eI/ vowel).

Data Processing: Acoustic data were forced-aligned and formants (F1, F2) extracted every 2ms, then normalized using the $\Delta$ F method to account for vocal tract length. Ultrasound tongue contours were automatically labeled using DeepLabCut (DLC), identifying 11 anatomical landmarks.
Articulatory Quantification: To isolate speaker-specific strategy from token-level variability, the authors applied Generalized Procrustes Analysis (GPA) to tongue contours across all vowels, followed by Principal Component Analysis (PCA) in tangent space. This yielded three principal components (PCs) representing modes of tongue shape variation:
- PC1: Dorsal convexity (doming).
- PC2: Anterior dorsal convexity and tongue root advancement/retraction.
- PC3: Mid-tongue concavity/convexity.
- For each speaker, the mean scores of these PCs were calculated over the initial 80% of the /i/ vowel (in the word bead) to serve as a stable predictor of articulatory strategy.
Statistical Modeling: Generalized Additive Mixed Models (GAMMs) were employed to model normalized F1 and F2 trajectories for the I-diphthong series (bead, bade, bide, buoyed). The models included smooth terms for time and duration, with the speaker's mean /i/ PC scores (i-PC1, i-PC2, i-PC3) and their interactions with time as predictors. Random effects accounted for speaker and token variability. Model significance was determined via likelihood ratio tests comparing full models against nested models excluding specific PC predictors.

Key Results

Strategy-Formant Link: The study confirms that articulatory strategy, proxied by the mean tongue shape in /i/, significantly predicts formant dynamics in I-diphthongs.
- i-PC1 (Dorsal Doming): Significantly predicted F1 trajectories for bade, bide, and buoyed, and F2 for bead and buoyed. Speakers with higher PC1 scores (more dorsal doming) exhibited earlier acoustic transitions.
- i-PC2 (Anteriority/Root): Significantly predicted F1 and F2 trajectories across all four vowels. Speakers with lower PC2 scores (more anterior constriction and tongue root advancement) produced earlier and faster formant transitions.
- i-PC3 (Mid-tongue shape): Showed significant effects on F1 for all vowels and F2 for bead and buoyed, though the effects were generally smaller in magnitude.
Mechanism of Variation: The observed acoustic differences are not artifacts of vowel duration (which showed no significant correlation with PC scores). Instead, the results suggest a kinematic explanation: speakers with specific tongue shapes (e.g., highly domed or anteriorly constricted) must traverse greater articulatory distances to reach the offglide target. To achieve this within a fixed temporal window, these speakers must employ higher articulatory velocities. This results in steeper and earlier formant transitions, consistent with linear harmonic oscillator models of speech movement.
Vowel-Specificity: The acoustic effects of articulatory strategy were most robust in diphthongs with larger inherent changes (e.g., bade, bide, buoyed) and less pronounced in bead (which has a smaller articulatory displacement), supporting the view that dynamic variability is more observable when articulatory displacement is large.

Significance and Claims
The paper claims to provide empirical corroboration for the hypothesis that articulatory strategy is a primary source of individual variation in acoustic vowel dynamics. By linking specific tongue shape features in a static vowel to dynamic formant trajectories in diphthongs, the study bridges the gap between articulatory compensation (often viewed as a mechanism for acoustic invariance) and acoustic individuality.

The authors argue that while speakers compensate for vocal tract morphology to achieve stable acoustic targets, the process of compensation—specifically the magnitude of articulatory displacement required—imposes dynamic constraints. Greater displacement necessitates higher velocities, which in turn produces distinct, speaker-specific formant trajectories. This finding refines the understanding of speaker individuality, suggesting it arises not just from static anatomical differences, but from the dynamic consequences of the strategies speakers adopt to overcome those anatomical constraints. The study concludes that these dynamic properties are recoverable in the acoustic signal, offering a potential avenue for forensic speaker identification, while noting that the relationship is likely governed by universal articulatory-acoustic principles rather than language-specific settings.