Speech
The tongue changes the shape of the space inside the mouth. Working with the lips, jaw, and palate, it helps form vowels and consonants. Try saying “ee” and “ah” slowly and notice how its position changes.
Tongue Map research platform
Mapping the atlas.
Discovering
functional units.
Functional units and the tongue atlas form the core of our research. We identify patterns of coordinated tissue motion and use a shared reference space to study how tongue organization relates to speech.
41 tongue, speech & tracking records62 unique publication records2011–2026 collection coverage
01 /02 /03 /
Discover functional units ↗
Explore coordinated patterns of tongue motion.
Build on the atlas ↗
Use a shared reference for anatomy and function.
Explore the literature ↗
Search by topic, author, year, or identifier.
Start here / Tongue anatomy & function
A small organ with a remarkable range of motion.
Your tongue helps you speak, move food, swallow, and experience taste. These everyday tasks depend on muscles working together with the brain and sensory nerves.
Navigate the tongue in 3D.
Rotate, zoom, open a cutaway, and select a label to explore anatomy and function.
Synthetic teaching model using the team's fiber simulator. Shapes and locations are schematic; some muscles are described but not rendered separately.
Many muscles, one flexible structure.
The tongue is a muscular organ covered by a moist lining. Its tip and body lie toward the front of the mouth; its base extends toward the throat.
It has eight paired muscle groups. The intrinsic muscles lie within the tongue and help change its shape. The extrinsic muscles connect it to surrounding structures and help position it. Both groups work together during real movements.
Muscle fibers run in different directions and interweave. The tongue can lengthen, shorten, flatten, or bend without a rigid skeleton inside it. Its volume stays approximately constant as its shape changes, a property called a muscular hydrostat.
Eating & swallowing
During chewing, the tongue moves food between the teeth and mixes it with saliva. It gathers food into a small mass and pushes it toward the throat to begin swallowing.
Taste & sensation
Taste buds detect sweet, sour, salty, bitter, and umami. These tastes are detected across the tongue rather than in separate taste zones. Touch, temperature, and smell also contribute to the experience of food.
Why map tongue motion?
Different muscles can contribute to the same movement, and parts of the same muscle can behave differently. Looking at anatomy alone does not fully explain how the tongue performs a task.
An atlas is a shared reference map that helps researchers compare anatomy and motion across people. Functional units are patterns of tissue that move together during a task. They can cross muscle boundaries. Studying both can help us understand speech and investigate changes after disease or surgery.
Learn more: Tongue anatomy · NCBI Bookshelf · How taste works · NIH/NIDCD · Tongue muscle structure · Stone, Woo and colleagues.
Core research themes
Functional units and the tongue atlas.
How is tongue motion organized during speech, and how can that organization be compared across speakers? Functional units and a shared atlas address these complementary questions.
FUNCTIONAL ORGANIZATION
Identify coordinated patterns of motion.
Functional units describe spatial patterns of tongue tissue that move together during speech, with temporal components describing how those patterns change over time. Our work investigates shared and individual patterns and their relationship to speech acoustics.
These units are inferred from imaging-based motion representations. Their boundaries need not coincide with individual anatomical muscles.
ATLAS & COMPARISON
Place anatomy and function in a shared reference.
The atlas provides a common coordinate system for aligning tongue anatomy and motion across speakers. It supports comparison of functional organization while preserving the distinction between shared patterns and individual variation.
Together, the atlas and functional-unit analysis provide a framework for investigating tongue coordination, muscle mechanics, and speech.
Atlas & functional units
Highlights.
Functional units and the tongue atlas: selected figures and findings from published studies.
FUNCTIONAL UNITS
Functional units of tongue motion
The published figure compares two-unit representations of common and individual functional organization during the speech transitions in “a souk.” The study jointly identifies common and subject-specific units in an atlas reference space.
FOUNDATIONAL STUDY / 2019
A Sparse Non-Negative Matrix Factorization Framework for Identifying Functional Units of Tongue Behavior From MRI ↗
This earlier framework used tagged-MRI motion features, sparse non-negative matrix factorization, a probabilistic graphical model, and spectral clustering to identify subject-specific functional units. Evaluation included synthetic data and in vivo protrusion and speech tasks.
J. Woo et al. · IEEE Transactions on Medical Imaging 38(3), 730–740 (2019; online 2018)
View Figure 6: functional units during /s/ → /u/ ↗The later 2021 study jointly identifies common and individual units in an atlas reference space.
J. Woo and colleagues. Published Figure 8; original article rights apply.
J. Woo et al. · Medical Image Analysis 72, 102131 (2021) ↗The source image could not be loaded.
View this figure in the paper ↗TONGUE ATLAS
Atlas and variation during /ə/
The spatiotemporal atlas represents tongue shape and its variation across speakers. This published panel shows principal-component variation during /ə/, with MRI volumes above and transverse and genioglossus muscle renderings below.
J. Woo, Fangxu Xing, Junghoon Lee, Maureen Stone and Jerry L. Prince. Published Figure 7; original article rights apply.
J. Woo et al. · Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization 6(5), 520–531 (2018) ↗The source image could not be loaded.
View this figure in the paper ↗TONGUE ATLAS
Atlas and variation during /s/
The companion atlas panel shows principal-component variation during /s/. MRI volumes and corresponding muscle renderings illustrate differences in tongue shape within a common reference framework.
J. Woo, Fangxu Xing, Junghoon Lee, Maureen Stone and Jerry L. Prince. Published Figure 10; original article rights apply.
J. Woo et al. · Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization 6(5), 520–531 (2018) ↗ATLAS MAPS / FEATURED PAPER
Speech Map: A Statistical Multimodal Atlas of 4D Tongue Motion During Speech from Tagged and Cine MR Images ↗
Cine MRI supplies a shared anatomical reference; tagged MRI provides internal tissue motion. Speech Map combines these to compare motion and strain across speakers and characterize variation.
J. Woo et al. · Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization 7(4), 361–373 (2019; online 2017) · DOI: 10.1080/21681163.2017.1382393
Figure 4 · Reference atlas and speech motion
If the repository blocks the embedded view, open the original figure below.
Open Figure 4 ↗Explore all eight published figures
- Figure 1 · Cine and tagged MRI examples ↗
- Figure 2 · Speech Map workflow ↗
- Figure 3 · Mapping motion into atlas space ↗
- Figure 4 · Reference atlas and average motion ↗
- Figure 5 · Motion and strain in subject space ↗
- Figure 6 · Motion and strain in atlas space ↗
- Figure 7 · Patient and control strain comparison ↗
- Figure 8 · Principal components of motion variation ↗
Atlas images retain the original figure panels. The functional-unit highlight links to the published figure; its standalone image is not yet available here. Source links provide the full captions and study context.
Functional organization & supporting methods
From atlas and functional units to speech.
Begin with functional-unit discovery, then explore the anatomy, strain measurements, motion tracking, and synthesis methods that support the atlas and functional organization of speech. Each card includes a visual, summary, and publication link.
FUNCTIONAL ORGANIZATION / 2021
A deep joint sparse non-negative matrix factorization framework for identifying the common and subject-specific functional units of tongue motion during speech
Deep joint sparse non-negative matrix factorization identified tongue-motion components shared across speakers and components specific to individual speakers. The resulting functional units improved interpretability and reduced variation in unit size. The approach provides a way to compare coordinated tongue motion across individuals.
TONGUE MUSCLE ANATOMY / 2018
Structure and variability in human tongue muscle anatomy
High-resolution 3D MRI from 14 people was used to measure tongue muscle volumes, relationships, and variation across individuals. The study distinguished functional muscle volume—the region served by a muscle—from structural volume, which accounts for overlapping fibers. This anatomical framework helps interpret complex tongue organization and supports more realistic tongue models. These functional volumes are distinct from the motion-derived functional units described in other studies.
Explore published anatomy figures
PEDIATRIC VOCAL TRACT / 2026
Multi-Atlas-Based Segmentation of Pediatric Vocal Tract Anatomy in Dynamic Magnetic Resonance Imaging
This study compared multi-atlas approaches for segmenting pediatric vocal-tract anatomy in dynamic MRI. Corrective learning improved the reported average Dice score from 0.89 with 10 temporal frames to 0.92 with 40 frames. The result supports using temporal information to help delineate speech-related structures in small pediatric datasets.
MUSCLE MECHANICS / 2023
Analysis of Tongue Muscle Strain During Speech From Multimodal Magnetic Resonance Imaging
The authors combined diffusion MRI fiber orientations with tagged-MRI motion to estimate strain along tongue muscle fibers during speech. In eight healthy controls and two glossectomy patients, the analysis revealed differences in muscle-strain patterns. These measurements describe tissue deformation and do not directly measure muscle activation.
Figure context & method diagram
Structural MRI and labeled three-dimensional muscle reconstructions show the atlas and two glossectomy cases, including reconstructed flap regions. The original anatomical labels and orientation markers are retained. Published figure, served by PubMed Central. Credit: Shao and colleagues; original article rights apply. Source and full caption ↗
GENERATIVE SPEECH IMAGING / 2026
A speech-to-video synthesis approach using spatio-temporal diffusion for vocal tract MRI
This framework conditions a spatiotemporal diffusion model on speech audio to generate vocal-tract MRI sequences. Evaluation included healthy speakers and tongue-cancer patients, with reported adaptation to new speech inputs and favorable human evaluations. The original preprint comparisons below allow readers to inspect generated frames beside recorded MRI.
Explore original paper figures
ORIGINAL PAPER FIGURES
Speech-conditioned MRI: compare model outputs
Compare generated MRI frames with the recorded reference for a speaker saying “When the.” Read across time and down the model rows to inspect anatomical detail and motion consistency.
Look for continuity of the tongue contour across adjacent frames.
Pérez-Toro et al. · author preprint (2025), Figure 4, PDF page 15 · CC BY-NC-ND 4.0. Full source page, unchanged.
Open original figure page ↗PDF preview depends on your browser. Use the source link if it does not appear.
MRI SEQUENCE SYNTHESIS / 2024
Tagged-to-Cine MRI Sequence Synthesis via Light Spatial-Temporal Transformer
LiST² uses neighboring tagged-MRI frames to synthesize cine MRI, while recurrent sliding fine-tuning encourages consistency between overlapping sequence sections. The study reported improvements over the evaluated alternatives in image quality and temporal consistency. Original MRI comparisons and parameter-sensitivity plots are available below.
Explore original paper figures
ORIGINAL PAPER FIGURES
Tagged-to-cine MRI: inspect temporal consistency
Original MRI comparisons show LiST² + ReST, competing methods, and model ablations across consecutive sections. The panel supports inspection of anatomy and consistency between neighboring frames.
Compare frames across the section boundary for abrupt changes in tongue shape or texture.
Liu et al. · MICCAI (2024), Figure 2, PDF page 6. Original paper page, unchanged; article rights apply.
Open original figure page ↗PDF preview depends on your browser. Use the source link if it does not appear.
TISSUE MOTION TRACKING / 2026
DSHARP: Deep Incompressible Motion Estimation With Sinusoidal-Transformed Harmonic Phase for Tagged MRI
DSHARP combines harmonic-phase information with learned registration to estimate diffeomorphic, nearly incompressible tissue motion from tagged MRI. Evaluation included tongue, phantom, and cardiac data. The study reported improvements in tracking accuracy, computation speed, and preservation of incompressibility.
Each card pairs a study summary with a visual. Published images, redrawn results, and explanatory diagrams are labeled separately. Original PDF figure pages remain available in the expandable panels; standalone images are not yet available for every paper.
Code & demonstration data
Build on the atlas. Explore functional organization.
Atlas construction and segmentation resources provide a common anatomical reference for studies of motion and functional units. Explore the functional-unit study alongside these released tools.
ATLAS CONSTRUCTION
4D atlas construction code
Access the atlas-construction resource previously shared by the research team. Consult the repository for its files, instructions, version, and citation.
Resource link supplied by the teamDOI: 10.7910/DVN/OIULST
PEDIATRIC VOCAL-TRACT SEGMENTATION
Label fusion & corrective learning
Scripts and a small de-identified demonstration subset for atlas registration, label fusion, and corrective learning. The full pediatric dataset is restricted.
Code + demonstration subsetDOI: 10.7910/DVN/X8U4DY
Access and licensing are governed by each repository. These links do not imply release of all participant imaging or clinical data.
Publication library
Follow the evidence.
Explore atlas, functional-unit, and related imaging studies together in one publication collection. Search by author, topic, or year.
The collection brings together 62 unique publication records, including earlier atlas and functional-unit studies. Duplicate PMID records are merged; distinct conference and journal papers remain separate. Sources include the original publication export, PubMed metadata, and article acknowledgments.
Interactive synthetic model
Explore synthetic tongue fiber architecture.
Rotate the model, reveal muscle layers and cutaways, and explore volume-preserving deformation. Adjust the geometry and compare front, side, top, and oblique views.
Synthetic geometric model supplied by the research team. Fiber colors distinguish longitudinalis, transversus, and genioglossus. This is an explanatory simulation, not measured MRI or a participant-specific reconstruction.
About Tongue Map
Functional units and an atlas for speech imaging.
Tongue Map centers on two complementary themes: identifying functional units of coordinated tongue motion and constructing an atlas for comparison across speakers. Together, they connect anatomical organization, tissue mechanics, and speech acoustics.
PI: Jonghye Woo
Massachusetts General Hospital
Associate Professor of Radiology · Harvard Medical School
Core research theme
Our grant-supported research places functional units alongside atlas development at the center of understanding tongue organization and speech function.
Simulator contribution
The tongue simulator was created by Dr. Van Wedeen at Massachusetts General Hospital (MGH).
Student contributors
Zahra Prasad and Ellian Darlow helped develop the Tongue Map educational website, supporting the presentation of tongue anatomy, function, and research resources for students and the general public. We gratefully acknowledge their contributions.
Funding & acknowledgments
Supported by NIH awards R01DC018511, R01DE027989, R21DC016047, and K99/R00DC012575.
Explore with context.
What do the figures represent?
The Highlights section presents atlas figures and a functional-unit figure source from published studies, with author credits and source links. The findings section includes original source PDF figure pages and an anatomy image alongside a redrawn Dice chart and explanatory diagrams. The diagrams are clearly labeled and contain no patient MRI or simulated experimental results.
What is in the publication library?
There are 62 unique publication records in one combined collection. The default selection includes 41 tongue, speech, and tracking studies; the full collection includes broader imaging methods. Papers from all supported projects are mixed together and ordered by publication year.
How can I use the tongue simulator?
Drag to rotate and scroll or pinch to zoom. Toggle fiber layers and boundaries, select cutaways and camera presets, change geometry, and play or adjust the volume-preserving deformation. The simulator is a synthetic geometric model, not a participant-specific reconstruction.
How should resources be cited?
Use each publication’s PubMed record and each resource’s repository citation. The citation export contains the currently matching publication records in RIS format.