Automated Measurement of Geniohyoid Muscle Thickness During Speech Using Deep Learning and Ultrasound

Alisher Myrgyyassov; Bruce Xiao Wang; Yu Sun; Shuming Huang; Zhen Song; Min Ney Wong; Yongping Zheng

q-bio.QMFeb 26, 2026

Automated Measurement of Geniohyoid Muscle Thickness During Speech Using Deep Learning and Ultrasound

Authors: Alisher Myrgyyassov, Bruce Xiao Wang, Yu Sun, Shuming Huang, Zhen Song, Min Ney Wong, Yongping Zheng

Organizations: Biomedical Engineering Department, Hong Kong Polytechnic University, Hong Kong, China · English and Communication Department, Hong Kong Polytechnic University, Hong Kong, China · Research Institute for Smart Ageing, Hong Kong Polytechnic University, Hong Kong, China · Department of Language Science and Technology, Hong Kong Polytechnic University, Hong Kong, China

Abstract

Manual measurement of muscle morphology from ultrasound during speech is time-consuming and limits large-scale studies. We present SMMA, a fully automated framework that combines deep-learning segmentation with skeleton-based thickness quantification to analyze geniohyoid (GH) muscle dynamics. Validation demonstrates near-human-level accuracy (Dice = 0.9037, MAE = 0.53 mm, r = 0.901). Application to Cantonese vowel production (N = 11) reveals systematic patterns: /a:/ shows significantly greater GH thickness (7.29 mm) than /i:/ (5.95 mm, p < 0.001, Cohen's d > 1.3), suggesting greater GH activation during production of /a:/ than /i:/, consistent with its role in mandibular depression. Sex differences (5-8% greater in males) reflect anatomical scaling. SMMA achieves expert-validated accuracy while eliminating the need for manual annotation, enabling scalable investigations of speech motor control and objective assessment of speech and swallowing disorders.

Figures & tables

Figure 1: Visualization of the SMMA pipeline. Image (a) shows the original image and probe placement, (b) an automatically generated mask by UltraUNet, and (c) shows the middle 50% skeleton extracted from the mask and corresponding thickness measurements in pixels (px) and millimetres (mm).

Pair / Model	Dice (Mean $\pm$ SD)	IoU (Mean $\pm$ SD)
Ann1 vs Ann2	$0.9179\pm 0.0247$	$0.8493\pm 0.0410$
Ann1 vs Ann3	$0.9036\pm 0.0444$	$0.8270\pm 0.0682$
Ann2 vs Ann3	$0.9001\pm 0.0423$	$0.8209\pm 0.0654$
Attention UNet	$0.8707\pm 0.0332$	$0.7801\pm 0.0483$
DeepLab v3	$0.8792\pm 0.0175$	$0.7874\pm 0.0269$
SwinUNet	$0.8159\pm 0.0746$	$0.7006\pm 0.1046$

Table 1: Component 1 validation summary: human inter-annotator agreement and model performance (averaged across subjects and runs, with

\pm

indicating run-wise standard deviation).

Figure 2: Component 2 thickness measurement validation: SMMA automated measurements (left) vs. sonographer ground truth annotations (right).

Figure 3: Representative 30-second sample from the full recording of continuous muscle thickness tracking on a female subject during speech production. Subject produces repeated /a \textlengthmark / - /i \textlengthmark / - /u \textlengthmark / isolated vowel sequences (purple = /a \textlengthmark / , red = /i \textlengthmark / , green = /u \textlengthmark / , white = pause), each vowel lasting around 800 ms. Brief gaps in tracking occur when image quality temporarily degrades (e.g., 75s), demonstrating algorithm behavior under naturalistic recording conditions.

Vowel	Female (mm)	Male (mm)	$p$ -value
/a \textlengthmark /	$7.16\pm 0.73$	$7.45\pm 1.05$	0.200
/i \textlengthmark /	$5.76\pm 0.77$	$6.18\pm 0.87$	0.041*
/u \textlengthmark /	$6.41\pm 0.68$	$6.92\pm 0.84$	0.009**
* $p<0.05$ , ** $p<0.01$

Table 2: Geniohyoid muscle thickness across vowels and sex

Speech Production

See all 8 papers

Training-Free Cross-Lingual Dysarthria Severity Assessment via Phonological Subspace Analysis in Self-Supervised Speech Representations
Apr 11, 2026Bernard Muller, Antonio Armando Ortiz Barrañón, LaVonne RobertsDysarthria Assessment Clinical Speech Processing
The Scott-Morgan Foundation, Torquay, United Kingdom · Tecnológico de Monterrey, Monterrey, Mexico
Phonological Subspace Collapse Is Aetiology-Specific and Cross-Lingually Stable: Evidence from 3,374 Speakers
Apr 23, 2026Bernard Muller, Antonio Armando Ortiz Barrañón, LaVonne RobertsDysarthria Assessment
The Scott-Morgan Foundation, Torquay, United Kingdom · Tecnológico de Monterrey, Monterrey, Mexico
Laryngeal Structure Segmentation in High-Speed Videoendoscopy Using Deep Learning
Sep 22, 2026Sardar Nafis Bin Ali, Mohsen Zayernouri, Dimitar D. Deliyski +1Medical Image Segmentation Endoscopy
Department of Mechanical Engineering, Michigan State University, 428 S. Shaw Lane, East Lansing, 48824, MI, USA. · Department of Communicative Sciences and Disorders, Michigan State University, 1026 Red Cedar Rd, East Lansing, 48824, MI, USA.

Apr 11, 2026J/K move · Enter open · S save

Training-Free Cross-Lingual Dysarthria Severity Assessment via Phonological Subspace Analysis in Self-Supervised Speech Representations

Bernard Muller, Antonio Armando Ortiz Barrañón, LaVonne Roberts

The Scott-Morgan Foundation, Torquay, United Kingdom · Tecnológico de Monterrey, Monterrey, Mexico

Dysarthria Assessment Clinical Speech Processing Speech Quality Assessment

Open paper AskPDF arXiv

Dysarthric speech severity assessment typically requires trained clinicians or supervised models built from labelled pathological speech, limiting scalability across languages and clinical settings. We present a training-free method that quantifies dysarthria severity by measuring degradation in phonological feature subspaces within frozen HuBERT representations. No supervised severity model is trained; feature directions are estimated from healthy control speech using a pretrained forced aligner. For each speaker, we extract phone-level embeddings via Montreal Forced Aligner, compute d-prime scores along phonological contrast directions (nasality, voicing, stridency, sonorance, manner, and four vowel features) derived exclusively from healthy controls, and construct a 12-dimensional phonological profile.Evaluating 890 speakers across 10 corpora, 5 languages (English, Spanish, Dutch, Mandarin, French), and 3 primary aetiologies (Parkinson's disease, cerebral palsy, ALS), we find that all five consonant d-prime features correlate significantly with clinical severity (random-effects meta-analysis rho = -0.50 to -0.56, p < 2e-4; pooled Spearman rho = -0.47 to -0.55 with bootstrap 95% CIs not crossing zero). The effect replicates within individual corpora, survives FDR correction, and remains robust to leave-one-corpus-out removal and alignment quality controls. Nasality d-prime decreases monotonically from control to severe in 6 of 7 severity-graded corpora. Mann-Whitney U tests confirm that all 12 features distinguish controls from severely dysarthric speakers (p < 0.001).The method requires no dysarthric training data and applies to any language with an existing MFA acoustic model (currently 29 languages). We release the full pipeline and phone feature configurations for six languages.

Abstract

Speech Production

Explore similar work

Training-Free Cross-Lingual Dysarthria Severity Assessment via Phonological Subspace Analysis in Self-Supervised Speech Representations

Phonological Subspace Collapse Is Aetiology-Specific and Cross-Lingually Stable: Evidence from 3,374 Speakers

Laryngeal Structure Segmentation in High-Speed Videoendoscopy Using Deep Learning