[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-122866-en":3,"doc-seo-122866-105":30,"detail-sidebar-cat-0-en-105":90},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},122866,4398048949847,"Eliana","https://ap-avatar.wpscdn.com/avatar/400002536579ef2da7f?_k=1778318612642679267",8,"Research & Report","Discovering dynamical models of speech using physics-informed machine learning","Spoken language emerges from high-dimensional, time-varying physical movements whose dynamical principles remain a central question. This study applies physics-informed machine learning, specifically sparse symbolic regression, to discover new dynamical models of speech articulation. The approach is validated on simulated articulatory data, showing near-perfect recovery of the original dynamical model under strong variation in duration, initial conditions, target positions, and added noise. A proof-of-concept on empirical data yields a small set of candidate models spanning increasing complexity and accuracy.","Proceedings of the 13th International Seminar on Speech Production (2024)  \nDiscovering dynamical models of speech using physics-informed machine learning  \nSam Kirkham  \nLancaster University, UK [s.kirkham@lancaster.ac.uk](s.kirkham@lancaster.ac.uk)  \nAbstract  \nSpoken language is characterised by a high-dimensional and highly variable set of physical movements that unfold over time. What are the fundamental dynamical principles that underlie this signal? In this study, we demonstrate the use of physicsinformed machine learning (sparse symbolic regression) for discovering new dynamical models of speech articulation. We first demonstrate the model discovery procedure on simulated data and show that the algorithm is able to discover the original model with near-perfect accuracy, even when the data contain extensive variation in duration, initial conditions and target positions, as well as in the presence of added noise. We then demonstrate a proof-of-concept applying the same technique to empirical data, which reveals a small set of candidate dynamical models with increasing levels of complexity and accuracy.  \nKeywords: speech production, sparse symbolic regression, articulatory phonology, task dynamics, articulatory data  \n1. Introduction  \nA fundamental aim in the study of language is the discovery of abstract invariants that underlie the variability observed in performance. For example, speech production involves a set of low-dimensional combinatorial units that are physically realised as a set of variable and high-dimensional motions. How do we best model the relationship? One solution is proposed by Articulatory Phonology/ Task Dynamics (AP/TD), in which phonetics and phonology are isomorphic, with the fundamental unit being the speech gesture: an abstract goal-driven force directing the vocal tract to a target state (Browman and Goldstein 1992; Tilsen 2016; Iskarous 2017) .  \nSaltzman and Munhall (1989) propose a model of the gesture (hereafter abbreviated as SM89) as a critically damped harmonic oscillator (1), where k is a stiffness coefficient, m is amass coefficient, and the damping coefficient b = 2 √mk.  \nmx¨ + bx˙ + kx = 0 (1)  \nThe SM89 model has long been the core gestural equation underpinning AP/TD, but it fails to capture the quasisymmetrical velocity profiles and time-to-peak velocities typical of empirical data. Byrd and Saltzman (2003) show this can be solved via ramping functions, making gestural activation time-dependent. Sorensen and Gafos (2016) argue that this isan undesirable solution and that empirically realistic trajectories can be achieved by instead allowing the restoring force tobe non-linear via a cubic term dx3 in (2) . This also eliminates the need for time dependence once the gesture is initiated.  \nmx¨ + bx˙ + kx − dx3 = 0 (2)  \nThis model reproduces many characteristics of empirical velocity profiles, but there may still be some room for improvement. For instance, Elie, Lee, and Turk (2023) advance a general Tau model that outperforms the SG16 model in fitting empirical data. Beyond conventional models of the gesture, there is also considerable scope for further developing task dynamic models of other domains, such as prosodic time-series (Iskarous, Cole, and Steffman 2024), disordered speech (Parrellet al. 2023), and signed languages. In many cases, we might have a lot of data, but lack sufficient predictions of the underlying dynamics to propose a model, or we may seek alternative models that better fit empirical data. This raises a question: how can we efficiently develop new dynamical models of speech?  \nWe solve the problem of model discovery by leveraging recent developments in dynamical systems and machine learning that allow us to learn symbolic equations directly from data (Schmidt and Lipson 2009; Brunton, Proctor, and Kutz 2016) . In such cases, we want to find a small number of model terms that expose the underlying dynamics, as opposed to a neural network that may have a very","cbCaiu3ZBGeMxIZp","https://ap.wps.com/l/cbCaiu3ZBGeMxIZp","pdf",203882,1,4,"English","en",105,"# Abstract\n# Introduction\n# Methods","[{\"question\":\"What problem does the study address in speech research?\",\"answer\":\"It investigates the fundamental dynamical principles underlying the time-varying physical movements of speech and how to efficiently discover dynamical models from data.\"},{\"question\":\"How does sparse symbolic regression help discover speech articulation models?\",\"answer\":\"It learns symbolic equations directly from time-series data by selecting a small number of model terms from a candidate library, promoting sparsity to reduce overfitting.\"},{\"question\":\"What evidence does the study provide to validate the method?\",\"answer\":\"Experiments on simulated articulatory data show near-perfect recovery of the original model despite extensive variation and added noise, followed by a proof-of-concept on empirical data that produces candidate models with varying complexity and accuracy.\"}]","Discovering dynamical models of speech using physics-informed machine learning | PDF",1785813414,10,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":28},"discovering-dynamical-models-of-speech-using-physics-informed-machine-learning","",{"@graph":36,"@context":84},[37,53,67],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":21},"https://docshare.wps.com/document/discovering-dynamical-models-of-speech-using-physics-informed-machine-learning/122866/",{"url":52,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-08-04",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"What problem does the study address in speech research?","Question",{"text":74,"@type":75},"It investigates the fundamental dynamical principles underlying the time-varying physical movements of speech and how to efficiently discover dynamical models from data.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does sparse symbolic regression help discover speech articulation models?",{"text":79,"@type":75},"It learns symbolic equations directly from time-series data by selecting a small number of model terms from a candidate library, promoting sparsity to reduce overfitting.",{"name":81,"@type":72,"acceptedAnswer":82},"What evidence does the study provide to validate the method?",{"text":83,"@type":75},"Experiments on simulated articulatory data show near-perfect recovery of the original model despite extensive variation and added noise, followed by a proof-of-concept on empirical data that produces candidate models with varying complexity and accuracy.","https://schema.org",{"og:url":52,"og:type":86,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":88,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,119,122,127,130,133],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":21,"doc_module":4,"doc_module_name":46,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":46,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":46,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":46,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":46,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":46,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":29,"doc_module":4,"doc_module_name":46,"category_name":131,"show_sort_weight":29,"slug":132},"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":46,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]