德国 Carstens 三维电磁发音仪 AG501

摘要 · 中文整理：电磁发音仪（EMA）是一种可高精度捕捉舌、唇、下颌三维运动数据的仪器，已广泛用于言语产出研究、二语教学与临床康复。本研究介绍 SAIT-EMA 数据库，这是一个整合了汉语普通话母语者与来自不同母语背景的二语学习者的发音与声学数据的语料库。数据库包含18名被试（9男9女）的数据，其中6名二语学习者分别来自越南、秘鲁、墨西哥和俄罗斯，另有12名普通话母语者作为对照组（2名高水平者与10名一般母语者）。数据采用 Carstens AG501 EMA 系统采集，在发音器官上粘贴九个传感器：舌尖（TT）、舌叶（TB）、舌背（TD）、下切牙（LI）、上唇（UL）、下唇（LL）、鼻梁（NR）以及双侧乳突（LE 与 RE）。发音数据采样率为 250 Hz，同步音频采样率为 48 kHz，形成 .pos 格式的运动数据与标准 .wav 音频文件。语料选自汉语中介语语音语料库，覆盖音段、声调、变调与焦点韵律等广泛的语音现象，确保覆盖二语语音研究的关键问题。SAIT-EMA 数据库提供了专门针对不同语言背景汉语二语学习者发音模式的资源，为检验二语习得理论假设奠定了基础，并支持智能计算机辅助发音训练技术的发展。

原文摘要：Electromagnetic Articulography (EMA), an instrument to capture high-precision 3D movement data of the tongue, lips, and jaw, has been widely employed in speech production research, L2 instruction, and clinical rehabilitation. This study introduces the SAIT-EMA database, a Mandarin speech corpus that integrates articulatory and acoustic data from both native speakers of Mandarin and second language learners from diverse first-language backgrounds. The database comprises data from 18 participants (9 male and 9 female), including six L2 learners from Vietnam, Peru, Mexico, and Russia, and twelve native Mandarin speakers as a control group (two highly proficient speakers and ten general native speakers). Data were collected using the Carstens AG501 EMA system, with nine sensors attached to articulators: tongue tip (TT), tongue blade (TB), tongue dorsum (TD), lower incisors (LI), upper lip (UL), lower lip (LL), nasal bridge (NR), and bilateral mastoids (LE and RE). Articulatory data were sampled at 250 Hz, with synchronized audio recorded at 48 kHz, resulting in motion data in .pos format and standard .wav audio files. The speech materials were selected from a Mandarin interlanguage speech corpus, covering a broad range of phonetic phenomena including segments, tones, tone sandhi, and focus prosody—ensuring coverage of key issues in L2 phonetic research. The SAIT-EMA database offers the resource specifically targeting articulatory patterns in Mandarin L2 learners from diverse language backgrounds. It provides a valuable foundation for testing theoretical hypotheses in L2 acquisition and supports the development of intelligent computer aided pronunciation training technologies.

关键词：电磁发音仪测量、汉语普通话语料库、二语习得、发音—声学数据、计算机辅助发音训练

研究目标：解决缺乏涵盖普通话母语者与不同母语背景二语学习者的高精度三维发音运动学-声学同步数据库的问题，为二语习得理论检验与智能发音训练技术提供数据基础。

主要结论：SAIT-EMA 数据库的发音运动数据在辅音对之间表现出系统性、可重复的差异及同类内一致性，运动学可靠且适合定量比较与建模；跨语言分析显示越南语者发 /tsʰ/ 时舌形更前突、拱起，而普通话母语者舌面较平。

创新点：首个专门针对多语言背景普通话二语学习者的三维 EMA 数据库，语料系统覆盖音段、声调、连读变调与焦点韵律，并采用 Procrustes 归一化支持跨语言舌形比较。

后续建议：扩大二语学习者的母语类型与样本量以增强跨语言比较的统计效力；补充纵向学习数据以追踪发音习得过程；融合超声或 MRI 数据获取更完整的舌形信息；基于数据库开发发音-声学逆映射与自动发音评估模型；增加自发言语语料以提高生态效度

APA 引文：Jia, H., Xie, Y., Fang, Q., & Sun, Y. (2026). SAIT-EMA: A tridimensional electromagnetic articulography database for Mandarin with diverse language backgrounds. Scientific Data, 13, 899. https://doi.org/10.1038/s41597-026-07237-9

仪器引用：Data were collected using the Carstens AG501 EMA system, with nine sensors attached to articulators: tongue tip (TT), tongue blade (TB), tongue dorsum (TD), lower incisors (LI), upper lip (UL), lower lip (LL), nasal bridge (NR), and bilateral mastoids (LE and RE).

摘要 · 中文整理：本研究旨在考察帕金森病（PD）构音障碍者的发音运动幅度，并确定运动减少与感知构音受损严重度之间的关系。40 名 PD 患者与 40 名对照参与研究。根据感知评分，PD 组被分为有构音子系统受损组（PDdys+artic，n=18）与无受损组（PDdys−artic，n=22）。三维电磁发音仪记录了段落朗读时下颌、下唇、舌前部与舌后部的运动。对 stroke 幅度在多个幅度级上的频率分布进行了分析。结果表明，与对照相比，PDdys+artic 组产生显著更多的小幅度、更少的大幅度舌与下唇 stroke，而下颌差异不显著。较高幅度级的舌 stroke 计数与构音受损严重度评分存在显著负相关。运动减少在舌和唇上比在下颌上表现更明显，且只有舌的运动减少可能反映在严重度评分中。

原文摘要：The study aimed to examine articulatory movement amplitudes in individuals with Parkinson's disease (PD) dysarthria and determine relationships between hypokinesia and perceived articulatory impairment severity. Forty talkers with PD and 40 controls participated. Based on perceptual ratings, the PD group was divided into those with (PDdys+artic, n = 18) and without (PDdys−artic, n = 22) articulatory subsystem impairment. Three-dimensional electromagnetic articulography recorded jaw, lower lip, anterior tongue, and posterior tongue movements during paragraph reading. Stroke amplitudes were analyzed across frequency distributions at multiple amplitude levels. Results demonstrated that the PDdys+artic group produced significantly more small-amplitude and fewer large-amplitude tongue and lower lip strokes compared to controls, while jaw differences were nonsignificant. Significant inverse correlations existed between tongue stroke counts at higher amplitudes and articulatory impairment severity ratings. Hypokinesia manifests in the tongue and lip more so than in the jaw. Only hypokinesia of the tongue may be reflected in severity ratings.

关键词：帕金森病、构音障碍、运动减少、电磁发音仪、运动幅度

研究目标：明确帕金森病构音障碍者的运动减少（hypokinesia）在舌、下唇、下颌三个发音器官上的分布特征，以及这种运动幅度变化是否与临床感知的构音受损严重度相关。

主要结论：有构音受损的 PD 患者舌与下唇的小幅度 stroke 显著增多、大幅度 stroke 显著减少，而下颌无显著差异；只有舌的高幅度 stroke 频数与严重度评分显著负相关，提示舌的运动减少最具功能意义。

创新点：首次以 stroke 幅度频率分布（而非平均幅度）刻画 PD 构音障碍的运动减少，并按感知评分将患者分层，从而分离出与构音受损直接相关的发音器官级运动学特征。

后续建议：扩展至句子级与自发言语任务，检验幅度分布特征的任务稳定性；结合语速与响度调控任务考察运动减少的可塑性；将 stroke 幅度分布特征用于机器学习分类以辅助早期诊断；纵向随访以评估幅度分布特征对疾病进展的敏感性；增加口语清晰度测量以验证舌运动减少与可懂度的关系

APA 引文：Mefferd, A. S., Tjaden, K., Dietrich, M. S., & Brown, A. E. (2026). Amplitude distribution characteristics of the tongue, lower lip, and jaw in talkers with dysarthria due to Parkinson's disease. Journal of Speech, Language, and Hearing Research, 69(3), 874–890. https://doi.org/10.1044/2025_JSLHR-25-00078

仪器引用：three-dimensional electromagnetic articulograph (AG501, Carstens Medizintechnik) ... kinematic data were sampled at 250 Hz ... acoustic signal was sampled at 48 kHz

摘要 · 中文整理：言语运动本质上十分复杂，测量这些运动具有挑战性，通常需要在面部结构周围使用多个物理传感器。我们开发了一种整合面部关键点检测器与基于 Transformer 的神经网络模型 CoTracker 的新方法。测试表明其精密度达到亚毫米级（标准差约 0.15 mm），优于单独使用面部关键点检测（大于 0.3 mm），三维追踪性能与电磁发音仪相当（RMSE 约 0.3 mm）。该方法在成人与幼儿中表现同样良好。

原文摘要：Speech movements are inherently complex. Measuring these movements is challenging and often requires the use of multiple physical sensors around facial structures. We developed a novel approach that integrates a facial landmark detector and CoTracker, a transformer-based neural network model. Testing revealed submillimeter precision (approximately 0.15 mm standard deviation) superior to facial landmark detection alone (greater than 0.3 mm), with 3D tracking performance comparable to electromagnetic articulography (approximately 0.3 mm RMSE). The approach performed similarly well across adults and young children.

关键词：计算机视觉、面部关键点检测、无标记追踪、唇部运动学、言语产生

研究目标：解决传统言语运动测量依赖贴附在面部的物理传感器、难以用于幼儿与临床场景的问题，开发并验证一种基于普通视频的亚毫米精度无标记三维唇部运动追踪方法。

主要结论：面部关键点检测与 CoTracker 结合的无标记系统在头动条件下中位标准差低于 0.2 mm，相对 AG501 的 RMSE 多数低于 0.4 mm，成人与幼儿表现相当，精度接近高端有标记系统。

创新点：首次将 Transformer 点追踪模型 CoTracker 与面部关键点检测器结合用于三维言语运动追踪，并以 AG501 电磁发音仪为金标准在成人与幼儿中系统验证其亚毫米级精密度与准确度。

后续建议：扩展至舌与下颌等口腔内发音器官的间接估计；在构音障碍等临床人群中验证追踪鲁棒性；评估不同光照、相机分辨率与视角对精度的影响；实现实时处理以支持生物反馈训练；与声学信号融合构建多模态言语运动分析工具

APA 引文：Lovell, A., Liu, J., Borovsky, A., Yeh, R. A., & Kim, K. S. (2026). 3-D markerless tracking of speech movements with submillimeter accuracy. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 34, 2382–2393. https://doi.org/10.1109/TNSRE.2026.3688903

仪器引用：Participants were seated in the articulograph (AG501) with two sensors (black sensors with white wires) attached to their lips (one on the upper lip and the other on the lower lip). ... For this validation test, LA measured by AG501 was considered the 'ground truth,' given that it is arguably the most accurate speech kinematic recording system.

摘要 · 中文整理：本文设计并建立了一个包含发音运动学、声学、声门与面部微表情的普通话多模态情感语音数据库，并从语料设计、被试选择、录制细节与数据处理等方面进行了详细描述。各类信号同时标注离散情感标签（中性、高兴、愉悦、冷漠、愤怒、悲伤、悲痛）与维度情感标签（愉悦度、唤醒度、支配度）。本文通过对维度标注数据的统计分析验证了维度标注的有效性，并对标注者的SCL-90量表数据加以校验，结合PAD标注数据进行分析，以探究标注中的离群现象与标注者心理状态之间的内在关系。为验证数据库的语音质量与情感区分度，本文采用SVM、CNN和DNN三种基础模型计算七种情感的识别率。结果表明：仅使用声学数据时七种情感平均识别率约为82%；仅使用声门数据时平均识别率约为72%；仅使用运动学数据时平均识别率也达到55.7%。因此该数据库质量较高，可作为语音分析研究、尤其是多模态情感语音分析任务的重要数据来源。

原文摘要：A multi-modal emotional speech Mandarin database including articulatory kinematics, acoustics, glottal and facial micro-expressions is designed and established, which is described in detail from the aspects of corpus design, subject selection, recording details and data processing. Where signals are labeled with discrete emotion labels (neutral, happy, pleasant, indifferent, angry, sad, grief) and dimensional emotion labels (pleasure, arousal, dominance). In this paper, the validity of dimension annotation is verified by statistical analysis of dimension annotation data. The SCL-90 scale data of annotators are verified and combined with PAD annotation data for analysis, so as to explore the internal relationship between the outlier phenomenon in annotation and the psychological state of annotators. In order to verify the speech quality and emotion discrimination of the database, this paper uses 3 basic models of SVM, CNN and DNN to calculate the recognition rate of these seven emotions. The results show that the average recognition rate of seven emotions is about 82% when using acoustic data alone. When using glottal data alone, the average recognition rate is about 72%. Using kinematics data alone, the average recognition rate also reaches 55.7%. Therefore, the database is of high quality and can be used as an important source for speech analysis research, especially for the task of multimodal emotional speech analysis.

关键词：情感语音数据库、多模态情感识别、维度情感空间、三维电磁发音仪、电子声门仪

研究目标：解决普通话情感语音研究缺乏同步包含发音运动学、声学、声门与面部信息的多模态数据库的问题，并验证情绪标注的有效性与数据库的情绪区分能力。

主要结论：所建 STEM-E²VA 数据库质量较高：声学数据单独识别七类情绪的平均准确率约 82%，声门数据约 72%，运动学数据 55.7%；标注者心理状态（如抑郁、躯体化得分较高）与高愉悦度情绪的标注误差相关。

创新点：首次将 AG501 采集的发音运动学与声学、电声门图、面部视频同步集成于普通话情感语音数据库，采用自传式回忆与想象诱导获取自然情绪，并将标注者 SCL-90 心理量表与 PAD 标注结合分析标注离群现象。

后续建议：扩大被试规模并覆盖不同年龄与方言背景以提高泛化性；探索多模态特征融合模型以提升运动学模态的识别贡献；增加自发对话情绪语料以提升生态效度；发布标准化的数据划分与基线代码便于复现；引入连续情绪标注以支持时序情绪建模

APA 引文：Zhu, T., Li, L., Duan, S., Zhang, X., Xiao, Z., Jia, H., & Liang, H. (2024). Construction and evaluation of Mandarin multimodal emotional speech database. arXiv. https://doi.org/10.48550/arXiv.2401.07336

仪器引用：The study employed the EMA-AG501 (Electromagnetic Articulograph) to track speech production during emotional states. Sensors were positioned to track the tongue (T1-T3) and lips (L1-L4), totaling 13 sensors capturing detailed articulatory movement trajectories.