英国 Articulate Instruments 超声言语研究系统 MicrUs

MicrUs 超声言语研究系统（英国 Articulate Instruments Ltd）在售，由 MicrUs 扫描仪、探头、声卡、PStretch 单元、UltraFit 头戴装置与 AAA 软件构成。 探头置于下颌下方对舌部成像，记录原始扫描线数据；扫描仪每帧输出 TTL 脉冲，经 PStretch 展宽后与语音一同记录，实现毫秒级同步。 扫描仪 106x105x21 mm、0.26 kg，帧率最高 140 Hz；默认探头 MC4-2R20S-3 为 2-4 MHz、20 mm 半径。 用于典型言语与障碍言语的发音动力学、言语运动控制及视觉反馈治疗研究。 AAA 为标配，可提取舌轮廓与唇轮廓并导出数据；选配为 MC10-5R10S-3、MC8-4R20S-3、C5-2R60S-3 探头与唇部摄像套件。 厂商将其定位为研究设备，未给出监管分类。

英国 Articulate Instruments 超声言语研究系统 MicrUs

仪器信息

品牌Articulate Instruments
型号MicrUs
分类言语超声采集仪
仪器编码united-kingdom-articulate-instruments-ultrasound-speech-research-system-micrus
在售状态在售
地区来源英国
更新时间2026-10-10
研究领域语音学与言语科学、语音信号处理、司法语音学、言语运动控制与发音建模

技术参数

共 90 条规格记录

产品状态在售
产品线替代关系为已停产 EchoB 超声系统的现代等效机型
系统构成MicrUs 扫描主机（含硬件同步端口）、一个或多个探头、外置声卡（Focusrite Scarlett Solo 2 通道或 4i4 4 通道）、PStretch 脉冲展宽单元、UltraFit 头戴式探头固定装置
可选探头型号微凸阵 5–10 MHz、10 mm 半径探头；凸阵 4–8 MHz、20 mm 半径探头；凸阵 2–4 MHz、60 mm 半径探头
唇部视频采集配置头戴式摄像机支架（前向或侧向）含 NTSC 摄像机、视频采集卡（USB2）、视频/音频同步单元
探头选型适用性MC4-2R20S-3（20 mm 半径、2–4 MHz）（成人及 6 岁以上儿童）
探头选型适用性适用于儿童及体型较小的成人；因最低频率 5 MHz 穿透深度较短，不适用于成人舌轮廓成像（MC10-5R10S-3 微凸阵探头）
探头选型适用性尺寸过大，不适用于言语研究（C5-2R60S-3 凸阵探头）
探头跨系统兼容性仅支持为本机型设计的探头；其他系统探头不可用
探头频率范围2–4 MHz（MC4-2R20S-3 凸阵探头（推荐默认探头））
探头频率范围5–10 MHz（MC10-5R10S-3 微凸阵探头）
探头频率范围5–8 MHz（10 mm 半径微凸阵探头；Rev Jan 2017 宣传册）
探头频率范围4–8 MHz（MC8-4R20S-3 凸阵探头）
探头频率范围2–5 MHz（C5-2R60S-3 凸阵探头）
探头曲率半径20 mm（MC4-2R20S-3 凸阵探头）
探头曲率半径10 mm（MC10-5R10S-3 微凸阵探头）
探头曲率半径60 mm（C5-2R60S-3 凸阵探头）
探头阵元数64
扫描线数64 或 128（100% 视场下的扫描线密度设置）
扫描深度60–230 mm（MC4-2R20S-3 凸阵探头）
扫描深度30–150 mm（MC10-5R10S-3 微凸阵探头）
扫描深度30–150 mm（MC8-4R20S-3 凸阵探头）
扫描深度90–230 mm（C5-2R60S-3 凸阵探头）
最大视场角136°（10 mm 半径、5–10 MHz 微凸阵探头（MC10-5R10S-3））
最大视场角101°（20 mm 半径、2–4 MHz 凸阵探头（MC4-2R20S-3））
最大视场角150°（10 mm 半径微凸阵探头；Rev Jan 2017 宣传册）
最大视场角92°（20 mm 半径、2–4 MHz 凸阵探头；Rev Jan 2017 宣传册）
帧率最高 140 Hz
帧率典型 119 Hz（70 mm 深度；110° FOV；10 mm 半径探头）
帧率典型 81 Hz（80 mm 深度；101.2° FOV；20 mm 半径探头）
帧率61–119 Hz（典型值；含同步端口与不含同步端口两种配置）
帧率典型 119 Hz（70 mm 深度；120° FOV；10 mm 半径探头；Rev Jan 2017 宣传册）
帧率典型 81 Hz（90 mm 深度；92° FOV；20 mm 半径探头；Rev Jan 2017 宣传册）
原始扫描线数据记录记录原始超声扫描线数据，而非插值后的图像数据
成像声速常数1540 m/s（组织内声速设计值；精度验证体模须按该声速校准）
硬件同步信号输出每帧成像完成时输出 TTL 脉冲
硬件同步端口可加装性初次采购后不可加装或拆除
同步信号输出接口位置主机后部最右侧金色接头
帧同步脉冲宽度数纳秒（经 PStretch 展宽前的原始帧闪脉冲）
同步脉冲展宽后宽度数毫秒（经 PStretch 单元展宽后）
同步信号输出电平线路电平（line level）（接 Focusrite Scarlett Solo Gen2/Gen3 的通道 2）
同步信号连接线缆SMA 转 BNC 电缆（主机至 PStretch Sync IN）；BNC 转 Jack 电缆（PStretch SYNC OUT 至 Scarlett LINE IN）
同步脉冲记录采样率22 kHz、44 kHz（帧闪脉冲经 PStretch 展宽后由声卡采集）
音频与图像同步精度毫秒级
可同步记录数据流语音音频、超声图像、唇部视频、EMA
录制时长数分钟（AAA 与 SonoSpeech 常规录制模式）
录制时长1 小时以上（受磁盘空间限制）（连续直写硬盘（direct-to-disk）录制模式）
供电方式USB3（或市电）
供电方式可由笔记本电脑电池供电用于野外工作
主机供电电压5 V（由笔记本电脑 USB 连接取电，通常无需外接电源）
脉冲展宽单元供电电压5 V（可由连接 PC 的 USB 集线器或市电适配器供给）
电源输入电压100–240 V（MicrUs 与唇摄像机的开关电源）
电源输入频率50–60 Hz（MicrUs 与唇摄像机的开关电源）
配套超声成像软件Echowave II
配套言语分析软件Articulate Assistant Advanced（AAA）
软件版本v222.0.0（2026 年 2 月 3 日发布的当前下载版本）
软件探头类型支持范围不支持（AAA 目前无法从线阵探头录制数据）
操作系统兼容性Windows 7、8、10、11（所有版本，32/64 位）（计算机推荐规格）
操作系统兼容性Windows XP SP3、Windows Vista SP2、Windows 7、8、10（所有版本，32/64 位）（计算机推荐规格；Rev Jan 2017 宣传册）
macOS 支持情况不支持；Mac 须引导进入 Windows 操作系统
计算机处理器要求i5/i7 1.8 GHz 或更快（计算机推荐规格）
计算机处理器要求i3/i5/i7 1.8 GHz 或更快（计算机推荐规格；Rev Jan 2017 宣传册）
计算机内存要求8 GB 或以上（计算机推荐规格）
计算机内存要求4 GB 或以上（计算机推荐规格；Rev Jan 2017 宣传册）
计算机屏幕分辨率要求1024×768 或以上（计算机推荐规格）
计算机 USB 端口要求2 个或更多 USB 3.x/USB-C 端口（可能需要集线器）（计算机推荐规格）
计算机 USB 端口要求2 个或更多 USB 3.0 接口（计算机推荐规格；Rev Jan 2017 宣传册）
系统所需 USB 端口数量4（用于 AAA 许可密钥、PStretch 5V 供电、MicrUs 与 Focusrite Scarlett 声卡）
系统所需 USB 端口数量5（加装唇视频摄像机选件时，增加 DFG2USB 视频采集卡）
外形尺寸106×105×21 mm
主机重量0.26 kg
整机重量0.6 kg（含一只探头）
唇部摄像机重量70 g（选配唇视频摄像机）
唇部摄像机信号制式NTSC 隔行模拟（选配唇视频摄像机）
唇部摄像机图像分辨率640×480（选配唇视频摄像机；原始隔行图像）
唇部摄像机图像分辨率640×240（选配唇视频摄像机；经 AAA 去隔行后）
唇部摄像机帧率30 fps（选配唇视频摄像机；原始隔行图像）
唇部摄像机帧率60 fps（选配唇视频摄像机；经 AAA 去隔行后）
头戴装置探头座兼容性随附两个探头座，分别适配 MC4-2R20S-3 与 MC10-5R10S-3 探头（UltraFit 头戴式探头固定装置）
运行噪声无风扇噪声
医疗器械认证状态不具 CE 与 FDA 认证（含同步端口的研究型配置）
医疗器械认证状态已获 CE 与 FDA 批准（不含同步端口的临床型配置）
探头认证状态仅 CE 认证（5–10 MHz、10 mm 半径微凸阵探头与 2–4 MHz、20 mm 半径凸阵探头）
探头认证状态FDA 与 CE 认证（4–8 MHz、20 mm 半径凸阵探头）
外表面清洁方法用蘸中性清洁剂的软布擦拭
禁用清洁剂含氯、氨、氟碳化合物或碳氢化合物的溶液
探头消毒用品Clinell Universal antibacterial alcohol-free wipes（擦拭或喷洒）；ENZOL™ Enzymatic Detergent (Cidezyme)、MetriZyme™、Klenzyme® Enzymatic Instrument Detergent（无醇清洗）
探头禁用清洁剂异丙醇（会被硅胶吸收并使其软化受损）
超声耦合剂Parker Aquasonic 100 等经认可的超声耦合剂
适用研究对象典型言语与言语障碍的发音动力学研究

相关实验室

以下机构来自本仪器关联论文的作者署名，不代表已确认采购或持有该仪器。

兰卡斯特大学

地区：英国

署名作者：萨姆·柯卡姆、贾斯汀·J·H·罗

研究方向：言语运动控制、动力学模型、超声舌部成像、电磁发音仪

关联论文 2 篇

曼彻斯特大学

地区：英国

署名作者：帕特里齐娅·斯特雷哈尔丘克

研究方向：言语运动控制、动力学模型、超声舌部成像、电磁发音仪

关联论文 1 篇

MTA-ELTE Lendület舌发音研究组

地区：匈牙利

署名作者：亚历山德拉·马尔科

研究方向：语音处理、超声成像、深度学习

关联论文 1 篇

布达佩斯技术与经济大学

地区：匈牙利

署名作者：塔马什·加博尔·恰波

研究方向：语音处理、超声成像、深度学习

关联论文 1 篇

实验方法

研究方向：正中矢状面舌运动的超声舌位成像、发音-声学映射与静默语音接口、说话人特异性发音策略与说话人区分、发音运动学的动力学（任务动力学）建模、基于深度学习的舌廓解剖标志点追踪

基于Micro超声舌位图像的残差网络发音-声学映射实验

方法说明：使用Articulate Instruments公司的Micro言语超声系统在正中矢状面记录说话人朗读匈牙利语句子时的舌运动（81.67 fps），同步记录语音信号；比较原始扫描线（raw scanline）超声数据与经AAA软件插值参数转换得到的扇形（wedge）图像作为残差网络输入时的发音-声学映射效果，并探索不同输入像素尺寸对语音合成质量的影响，为静默语音接口寻找最优的超声图像表示。

实验仪器：Micro言语超声采集系统 Micro；凸阵超声探头 2–4 MHz/64阵元/20 mm曲率半径；电容麦克风 TG H56c；Articulate Assistant Advanced软件 AAA V219.08

实验步骤

  1. 数据采集

    说明：使用Micro超声系统与2–4 MHz/64阵元、20 mm曲率半径的凸阵探头在正中矢状面记录舌运动，帧率81.67 fps；同时用Beyerdynamic TG H56c电容麦克风以22 050 Hz采样记录语音。

    关键点：探头位置固定于下颌下方，超声与音频通过自定义工具实现同步

  2. 语料朗读

    说明：多名匈牙利语男女说话人各朗读209个句子，本研究选取一名女性说话人（#048）的数据进行实验。

    关键点：语料需覆盖足够的音素与发音动作变化

  3. 图像表示转换

    说明：利用AAA软件（V219.08）导出的插值参数，将64×842的原始扫描线数据转换为扇形超声图像；再将两类图像缩放为多种像素尺寸（如64×43、64×64）。

    关键点：扫描线到扇形图像的几何变换须使用AAA记录的探头参数

  4. 网络训练与评估

    说明：以残差网络（ResNet）作为发音-声学映射模型，输入超声图像、输出声码器谱参数，比较不同图像表示和尺寸下的合成质量。

    关键点：以客观指标（如MSE、MCD）与主观听感评价综合衡量

实验数据

正中矢状面超声舌位图像序列

数据来源：Micro超声系统在朗读匈牙利语句子时以81.67 fps采集

实验样本：原始扫描线数据64×842像素，转换后扇形图像缩放为64×64像素

数据呈现：扇形B模式超声图像显示、扫描线灰度图显示

同步语音信号

数据来源：Beyerdynamic TG H56c电容麦克风以22 050 Hz录制

实验样本：与超声帧同步的语音波形及其声码器谱参数

数据呈现：波形图与频谱图

方法关联论文：面向基于残差网络的发音-声学映射优化超声舌位图像表示

基于超声舌形特征与似然比检验的说话人区分实验

方法说明：对40名英格兰西北部英语说话人使用EchoB与Telemed Micro言语超声系统采集元音发音时的正中矢状面舌位超声数据，采用Articulate Assistant Advanced中集成的DeepLabCut自动标注舌廓11个解剖标志点；对舌轮廓进行广义普氏分析（GPA）与主成分分析，得到舌尺寸与舌形状特征，并在似然比框架下以等错误率（EER）和Cllr评估各特征的说话人区分能力。

实验仪器：Telemed Micro言语超声采集系统 Micro；EchoB超声系统 EchoB；探头稳定头盔；Articulate Assistant Advanced软件 AAA 2.20

实验步骤

  1. 参与者与语料设计

    说明：招募40名英格兰西北部（大曼彻斯特与兰开夏）英语说话人，在固定音段环境（/bV/、/bVd/，如bee、bead、bar、bard）中朗读全套重读元音，嵌入载体句，每项刺激重复4–6次。

    关键点：刺激从屏幕呈现，要求尽量自然朗读

  2. 超声数据采集

    说明：10名说话人使用EchoB系统、30名说话人使用Articulate Instruments的Telemed Micro言语超声系统采集正中矢状面舌位图像，帧率59.5–101 fps（中位数81.3 fps），使用头戴式稳定装置减少探头移动。

    关键点：探头稳定头盔保证探头相对头部位置固定

  3. 舌廓自动标注

    说明：在Articulate Assistant Advanced 2.20中使用DeepLabCut（预训练MobileNet1.0模型v1.1.0）自动标注舌廓11个解剖标志点，导出为以mm为单位的笛卡尔坐标并旋转到咬合平面。

    关键点：以说话人中位标志点为参照剔除偏差超过3.5个中位绝对偏差的异常试次

  4. 形状分析

    说明：用R的shapes包对舌廓做部分GPA（保留尺寸与形状）和完全GPA（仅保留形状），随后在切空间做主成分分析，提取前三个主成分。

    关键点：区分尺寸-形状分析与纯形状分析

  5. 似然比检验

    说明：用fvclrr包进行交叉验证的似然比检验，多元核密度公式计算得分，逻辑回归校准后以EER和Cllr评估各主成分及其组合的说话人区分性能。

    关键点：同一说话人数据分半用于同说话人比较，共产生1600个成对比较得分

实验数据

正中矢状面舌廓标志点坐标

数据来源：Micro/EchoB超声图像经DeepLabCut自动标注导出

实验样本：每个试次元音声学中点处11个标志点的（x, y）坐标（mm），共10 407个试次

数据呈现：舌廓样条曲线图、主成分效应图（均值形状±3 SD）

说话人区分性能指标

数据来源：似然比交叉验证检验结果

实验样本：尺寸-形状PC1+3系统EER为5.6%、Cllr为0.213

数据呈现：EER–Cllr散点图、性能表格

方法关联论文：作为说话人区分依据的元音产出发音策略

超声与电磁发音仪同步采集下的发音动力学模型参数估计实验

方法说明：对6名北部英式英语女性说话人同步采集电磁发音仪（EMA）与超声舌位数据：超声在Articulate Assistant Advanced中以约81 Hz用Telemed MicrUS（Micro）扫描仪及20 mm曲率半径、64阵元、2 MHz探头记录，EMA用Carstens AG501以1250 Hz记录；通过TTL脉冲实现音频、EMA与超声的时间同步；利用DeepLabCut追踪超声图像中的舌廓解剖标志点，将EMA降采样并与超声轨迹对齐，估计线性谐振子（去除临界阻尼约束）的刚度与阻尼参数，比较两种模态得到的动力学参数是否一致。

实验仪器：Telemed MicrUS（Micro）言语超声扫描仪 MicrUS；凸阵超声探头 20 mm曲率半径/64阵元/2 MHz；电磁发音仪 AG501；麦克风 Opus 55；麦克风前置放大器 m101；Articulate Assistant Advanced软件

实验步骤

  1. 参与者与语料

    说明：6名北部英式英语女性说话人，在She said X与She said X eagerly两种载体句中产出/bV/与/bVd/语境下的全部英式英语元音，每人4次重复（1人5次），共分析1095个词例。

    关键点：因探头移动过大剔除2名说话人的部分区块

  2. 仪器同步采集

    说明：EMA用Carstens AG501在舌尖、舌中、舌背、上下唇、下齿放置传感器并以1250 Hz记录；超声在AAA中以约81 Hz用Telemed MicrUS扫描仪（20 mm半径、64阵元、2 MHz探头）记录并以头盔稳定探头；音频用Beyerdynamic Opus 55麦克风经Grace Designs m101前置放大以48 kHz录制。

    关键点：每次刺激呈现时触发TTL脉冲并记录到各系统以实现时间对齐

  3. 数据处理

    说明：用Montreal Forced Aligner强制对齐并人工校正；用DeepLabCut预训练舌模型追踪超声图像11个舌标志点及短腱、舌骨点，导出mm坐标并按咬合平面旋转；EMA降采样到超声帧率，两者中心化后用5阶离散余弦变换平滑。

    关键点：咬合平面通过每位说话人的咬板记录估计

  4. 动力学参数估计

    说明：对CV段的舌位轨迹拟合无临界阻尼约束的线性谐振子模型，估计刚度与阻尼参数，比较超声与EMA得到的参数，并评估下颌短腱追踪对下颌运动的捕捉。

    关键点：放宽临界阻尼约束以获得更符合实测轨迹的自治模型

实验数据

超声舌廓标志点运动轨迹

数据来源：Telemed MicrUS超声在AAA中以约81 Hz记录，经DeepLabCut追踪

实验样本：11个舌标志点及短腱、舌骨点的x/y坐标时间序列（mm）

数据呈现：位置-时间曲线、与EMA轨迹叠加图

EMA传感器轨迹

数据来源：Carstens AG501以1250 Hz记录

实验样本：舌尖、舌中、舌背、唇及下齿传感器坐标

数据呈现：位置-时间曲线

方法关联论文：从超声舌部运动学估计动力学模型参数

主要部件

暂未收录主要部件

相关论文

共 3 篇；当前显示第 1 页

从超声舌部运动学估计动力学模型参数

原文题名：Dynamical model parameters from ultrasound tongue kinematics

发表年份：2025

作者：萨姆·柯卡姆；帕特里齐娅·斯特雷哈尔丘克

刊源：JASA Express Letters

DOI 10.1121/10.0039769
查看论文摘要与书目信息

摘要 · 中文整理：言语控制可以建模为一个动力学系统，其中发音器官被驱动趋向目标位置。此类模型通常用电磁发音仪（EMA）等肉点数据来检验，但近年的方法学进展使超声成像成为一种颇具前景的替代手段。本文评估能否从超声舌部运动学中可靠地估计线性谐振子模型的参数，并与同步记录的 EMA 数据所估参数进行比较。结果发现超声与 EMA 给出的动力学参数具有可比性，同时对下颌短肌腱的追踪也能充分刻画下颌运动。这支持使用超声运动学来评估动力学发音模型。

原文摘要：The control of speech can be modelled as a dynamical system in which articulators are driven toward target positions. These models are typically evaluated using fleshpoint data, such as electromagnetic articulography (EMA), but recent methodological advances make ultrasound imaging a promising alternative. We evaluate whether the parameters of a linear harmonic oscillator can be reliably estimated from ultrasound tongue kinematics and compare these with parameters estimated from simultaneously-recorded EMA data. We find that ultrasound and EMA yield comparable dynamical parameters, while mandibular short tendon tracking also adequately captures jaw motion. This supports using ultrasound kinematics to evaluate dynamical articulatory models.

关键词：言语运动控制、动力学模型、超声舌部成像、电磁发音仪、线性谐振子

研究目标：检验能否从超声舌位运动学数据可靠估计线性谐振子动力学模型参数，并验证其与同步EMA数据估计结果的一致性。

主要结论：超声与EMA得到的动力学参数相当，下颌短腱追踪能够充分捕捉下颌运动，支持用更廉价、无创且便携的超声运动学数据评估发音动力学模型。

创新点：首次利用同步采集的超声与EMA数据直接比较动力学模型参数估计，并采用放宽临界阻尼约束的简单线性谐振子模型。

后续建议：扩展到更多说话人、男性说话人及不同语言以验证泛化性；采用非线性或非自治模型进一步比较两种模态；量化EMA传感器位置与DeepLabCut标志点的空间对应关系；应用于临床与田野数据检验实际可用性

APA 引文：Kirkham, S., & Strycharczuk, P. (2025). Dynamical model parameters from ultrasound tongue kinematics. JASA Express Letters, 5(11), 115201. https://doi.org/10.1121/10.0039769

仪器引用：Ultrasound data were recorded in Articulate Assistant Advanced (Wrench, 2022) at ∼81 Hz using a Telemed MicrUS scanner with a 20-mm radius, 64-element, 2-MHz probe. The ultrasound probe was stabilised using a headset (Spreafico et al., 2018).

作为说话人区分依据的元音产出发音策略

原文题名：Articulatory Strategy in Vowel Production as a Basis for Speaker Discrimination

发表年份：2025

作者：贾斯汀·J·H·罗；帕特里齐娅·斯特雷哈尔丘克；萨姆·柯卡姆

刊源：Proceedings of Interspeech 2025

DOI 10.21437/Interspeech.2025-666
查看论文摘要与书目信息

摘要 · 中文整理：众所周知，说话人的发音方式因人而异，同时又受到解剖学与生物力学的约束。本研究提出的问题是：元音产出中的发音策略是否具有足够的说话人特异性，从而可以作为说话人区分的依据。我们对来自英格兰西北部 40 名英语说话人的舌形数据进行了广义 Procrustes 分析，并在似然比框架下评估了正交舌形特征的说话人区分潜力。结果显示，舌体大小是单项区分力最强的维度，而舌体较前部的形状变异总体上优于后部的形状变异。在特征组合的情形下，仅使用形状信息可以达到与同时使用大小加形状信息相当的说话人特异性水平，但前提是各特征之间不存在说话人层面的协变。

原文摘要：The way speakers articulate is well known to be variable across individuals while at the same time subject to anatomical and biomechanical constraints. In this study, we ask whether articulatory strategy in vowel production can be sufficiently speaker-specific to form the basis for speaker discrimination. We conducted Generalised Procrustes Analyses of tongue shape data from 40 English speakers from the North West of England, and assessed the speaker-discriminatory potential of orthogonal tongue shape features within the framework of likelihood ratios. Tongue size emerged as the individual dimension with the strongest discriminatory power, while tongue shape variation in the more anterior part of the tongue generally outperformed tongue shape variation in the posterior part. When considered in combination, shape-only information may offer comparable levels of speaker specificity to size-and-shape information, but only when features do not exhibit speaker-level co-variation.

关键词：说话人区分、司法声纹比对、超声舌位成像、元音发音、似然比、广义 Procrustes 分析

研究目标：探究元音发音中的舌形特征是否包含说话人特异性信息，以及哪些舌形特征对说话人区分最具信息量。

主要结论：所有舌形特征都携带一定说话人信息；舌尺寸的区分力最强，舌前部形状变异优于舌后部；尺寸-形状PC1+PC3组合表现最佳（EER 5.6%），纯形状特征组合可达到与之相当的水平，其性能与长时共振峰分布相当。

创新点：首次将似然比检验框架应用于发音（舌形）特征的说话人区分研究，并通过GPA分离舌尺寸与舌形状贡献。

后续建议：将舌形特征与声学特征（如共振峰）结合进行融合说话人比对；在不同录音时段采集数据以评估真实的说话人内变异；扩展到辅音及连续语流中的动态舌形特征；研究腭形态等解剖因素对发音策略个体差异的影响

APA 引文：Lo, J. J. H., Strycharczuk, P., & Kirkham, S. (2025). Articulatory strategy in vowel production as a basis for speaker discrimination. Proceedings of Interspeech 2025, 3504–3508. https://doi.org/10.21437/Interspeech.2025-666

仪器引用：The data were collected using the EchoB system (10 speakers) and the Telemed Micro ultrasound speech system (30 speakers) from Articulate Instruments Ltd. The sampling rate of the ultrasound system ranged between 59.5 and 101 frames per second (median = 81.3 fps). Headset stabilisation was used to minimise probe movement.

面向基于残差网络的发音-声学映射优化超声舌位图像表示

原文题名：Optimizing the Ultrasound Tongue Image Representation for Residual Network-Based Articulatory-to-Acoustic Mapping

发表年份：2022

作者：塔马什·加博尔·恰波；加博尔·戈斯托利亚；拉斯洛·托特；阿明·霍纳尔曼迪·尚迪兹；亚历山德拉·马尔科

刊源：Sensors

DOI 10.3390/s22228601
查看论文摘要与书目信息

摘要 · 中文整理：在语音处理领域，发音-声学映射（AAM）方法可以将超声舌位成像（UTI）作为输入。本研究比较了原始扫描线表示与经处理的扇形UTI图像作为残差网络输入用于AAM的效果。结果发现两种表示之间无显著差异，并确定原始扫描线输入的最优像素尺寸为64×43、扇形图像的最优尺寸为64×64，因此无需使用完整的64×842像素原始扫描线数据。更小的图像可以构建更小的网络，有利于静默语音接口应用，可帮助言语障碍者交流，并适用于军事场景或极端噪声环境。

原文摘要：Within speech processing, articulatory-to-acoustic mapping (AAM) methods can apply ultrasound tongue imaging (UTI) as an input. In this study we compared the raw scanline representation with the wedge-shaped processed UTI as the input for the residual network applied for AAM. We found no significant differences between the two representations, and identified the optimal pixel size to be 64 × 43 in the case of the raw scanline input, and 64 × 64 when transformed to a wedge. Therefore, it is not necessary to use the full original 64 × 842 pixels raw scanline. Smaller images allow the construction of smaller networks, which can be advantageous for silent speech interface applications, being helpful for the communication of the speaking-impaired, in military applications, or in extremely noisy conditions.

关键词：语音处理、超声成像、深度学习

研究目标：确定超声舌位图像作为深度残差网络发音-声学映射输入时的最优表示形式（原始扫描线还是扇形图像）与最优分辨率，以在不损失合成质量的前提下减小网络规模。

主要结论：原始扫描线与扇形图像两种表示在AAM性能上无显著差异；原始扫描线最优尺寸为64×43，扇形图像最优尺寸为64×64，无需使用完整的64×842原始数据，小尺寸输入可支持更紧凑的网络以用于静默语音接口。

创新点：首次系统比较了原始扫描线与扇形超声舌位图像在残差网络AAM中的效果，并给出了输入分辨率的最优选择依据。

后续建议：扩展到多说话人数据以验证结论的跨说话人泛化能力；引入时序建模（如3D卷积或Transformer）利用相邻帧信息；结合唇部视频等多模态输入提升合成质量；探索神经声码器以替代传统声码器提高听感自然度

APA 引文：Csapó, T. G., Gosztolya, G., Tóth, L., Shandiz, A. H., & Markó, A. (2022). Optimizing the ultrasound tongue image representation for residual network-based articulatory-to-acoustic mapping. Sensors, 22(22), 8601. https://doi.org/10.3390/s22228601

仪器引用：The tongue movement was recorded in midsagittal orientation using the 'Micro' ultrasound system (Articulate Instruments Ltd.) with a 2–4 MHz/64 element 20 mm radius convex ultrasound transducer at 81.67 fps.

使用资料与操作视频

Typical speech exemplar videos

简介：This video tutorial shows how to use the pre recorded video exemplars.As in the previous video,we will and select.An exemplar window by clicking on them or I can't button.It will always show the last exemplar that you had loaded.In this case.It was client specific exemplar.To load the prerecorded typical example,Our videos we click on the second button on the top,It will automatically show the video folder with a selection of videos.

发布时间：2020-02-14T11:09:53Z

Reward button

简介：This search tutorial video shows the use of the reward button,The reward button,is button with those wrapped present.Once you've completed a session,If you click this button and an animated GIF will appear on top of the outstanding display,It's randomly selected from a list of gifts and in a sub folder of the application folder.And it will be the gift will play for us as long as you and let it and to stop it playing.

发布时间：2020-02-14T10:17:36Z

Recalling client exemplar

简介：This small video shows how to use the Two Recall,An exemplar image for a client using the exemplar window for this possibly we don't need to have the prompt texts and the prompt window,So he can close those by clicking the buttons on the right hand side.But I've got a simple single display for the exemplar display.If we click on the dispily show,High example in a window.We'll get a second display here and not in the top.

发布时间：2020-02-14T09:14:14Z

Copy save

简介：This video tutorial shows how to use the exemplar window we can in life more click at the bottom right with a little clipboard，and the upside symbol to save the best effort by a client.So there we have clicked the button and it's immediately captured the image you can check the image to see whether it's suitable image to be stored.

发布时间：2020-02-14T07:55:40Z

Brightness and contrast

简介：This video shows how to adjust brightness and contrast and to find their controls for brightness and contrast.Click on the plus button at the bottom.This pops up a new set of buttons.If you click on the top one,Which is the outside settings button.It shows the brightness and contrast controls.If we slide the brightness,You can see goes dark too very light.If just contrast,Contrast,Increases or decreases.

发布时间：2020-02-14T07:11:11Z

Automatic palate trace

简介：In this tutorial,Video will show how to use the automatic palate trace function.Then open the mark overlay button.And if you click on that in this first Add palate,Symbol,You'll see these four extra symbols,And so we want to create a vocal tract.We'll click this first button.And then we need to say some sentences and for instance,Three young girls wore wool coats and boots in the car.And then we'll click on this third button to save it。

发布时间：2020-02-14T06:11:26Z

Image analysis tutorial 7 How to measure lip protrusion

简介：in the previous video we measured the lip distance along in this first vertical spline and I would quite like to measure the protrusion as well so we've selected this second chart here we'll add edit the analysis values and then we're going to create a new one and this is going to be and this again is going to be a 2d spline this time we just want to measure the position of this upper lip spline along in this measure

发布时间：2020-02-14T03:52:52Z

Automatic palate advanced

简介：This is short tutorial,Showing some more advanced features of the automatic palate tracing again clicking of the left button and again we can click the Create vocal tract button and we start Recording.Vocal track palate there.Sometimes you might get parts of the palate that extend beyond the region.

发布时间：2020-02-14T03:49:56Z

Image analysis tutorial 6 How to measure lip distance

简介：so now that we've added the template to lots of recordings I'll show you how we can actually make some measurements so if we go to a recording where I've already and applied the lip splines you can see they're moving through here if I just show you quickly three-headed splines

发布时间：2020-02-14T03:09:37Z

Add palate

简介：This is to tell you about adding an overlay or marker symbols.Press click the button on the second button down on the right.It will open up the mark window,We get selection of buttons,There look at Add palate choice,So click the first button,We can add a simple,Movable curve.Clicking the first button and you just click and drag the end points or the middle point.

发布时间：2020-02-14T02:50:22Z

Image analysis tutorial 5 Tools and methods for fitting 2D splines to lips

简介：this video shows some ways of manipulating moving the splines around changing the color so if we then go into headaches blinds in the splines itself we can select these and make the lines thicker for both lips and we can change the color as well and perhaps make not -for this your blue okay within the keyframes

发布时间：2020-02-14T02:45:23Z

Tutorial 10- Customizing the User Interface

简介：Hi and welcome to tutorial ten.Today we are going to be looking at the design feature of AAA to access design.You must right click on the top bar and click design.The design feature allows you to modify the layout of AAA into any preference which you wish.All of the current tabs along the top are presets,Which can be modified using the design function.Or you can create your own.

发布时间：2020-02-13T13:13:32Z

Tutorial 9- Batch Processing Splines

简介：And welcome to tutorial nine Batch Processing of splines.If you have not yet seen the tutorial on splines and other spline editing,Please see previous tutorial on it.So today we are going to set up splines for batch processing.As usual.We can go into the Edit splines dialogue to bring up,And similarly to how it is shown in the previous tutorial,We will go through that process.We will use the defaults which gives us these splines.

发布时间：2020-02-13T11:23:28Z

Tutorial 8- EPG and Formant Analyses in Publisher

简介：Hi and welcome to tutorial eight,This tutorial details the use of EPG data in publisher and also briefly looks at the use of wave data.This tutorial is a follow on from tutorial six,Which discusses the basics and the use of supplying data in publisher.It is recommended that you see that tutorial first.If you have not yet already.The graph that I have in front of me here is made from wave data.

发布时间：2020-02-13T10:09:01Z

Tutorial 7- Recording Ultrasound Data (Legacy)

简介：You ready to make a recording need to check that you've got the right line of text selected and a little of this one here,And then you got a clear image in the ultrasound window who just after speaker to say Ah,Ah,Can you say cat cat cat?We're already pressed the record button and then a screener again,And then our speaker will say the word.OK,Are you ready?

发布时间：2020-02-13T08:08:55Z

Tutorial 6- Publisher

简介：Publisher,Which is a tool that will allow you to export your data in a graphical format suitable for putting into documents and presentations.So to begin with,Publisher can be accessed in a number of different ways.The most simple is that it is under the file menu and publisher.You can import data from the track and you can also import it from the edit splines dialogue and the spline workspace.The spline workspace,You can import data using the button copy to publisher to do this.

发布时间：2020-02-13T06:44:44Z

Tutorial 5- The Spline Workspace

简介：This is tutorial five today we will be looking at the spline workspace.The spline workspace is a tool available in the software to allow you to manipulate splines you have placed and prepare them for publishing in scientific documents.The spline workspace is not available from the tabs at the top to open it,You have to right click the ultrasonic or video image,and click spline workspace.At the moment we can see that the workspace is empty.

发布时间：2020-02-13T03:54:19Z

Tutorial 3- Annotations and Splines

简介：And welcome to tutorial three.Today we will be looking at putting annotations on audio and putting spines on your ultrasonix images.So right now we are looking at the tab,Analyze Ultrasonix or Echo B.We had selected the data previously in the record Ultrasonixs ,Or Echo B tab.We are using the same demo data in this tutorial as we used in the previous.

发布时间：2020-02-13T01:36:00Z

Tutorial 4- Exporting data

简介：And welcome to tutorial four.Today we are going to look at how to export data in the two main ways.We will be using today the same data set as we had used in the previous tutorials.Both forms of exporting data that I will talk about today can be found under file in the top bar and export.The first we will look at is export files.This is a powerful tool to allow you to export your data in bulk from all your clients and recordings at once for import into other software.

发布时间：2020-02-12T11:31:12Z

Tutorial 2- User interface

简介：And welcome to tutorial too.Basic user interface,Today,I will show you the basics of the UI.When you first start up the software,You will get this loading screen,Just wait while it loads and it will take you into the main window of the software.When you first start,The window should look something like this for the purposes of this tutorial,We'll have a demo file loaded.

发布时间：2020-02-12T07:29:58Z

Tutorial 1- Installing and Uninstalling

简介：Hello and welcome to the setup tutorial.This tutorial covers how to install the software for articulate assistant advanced.There are two ways to install the software.First is from the CD,Which contains the files seen in this window.Another is to download the executable file from the Web Address Articulate Instruments.com as seen here.For installation,Please ensure that you have administrative privileges before beginning.We will now begin.

发布时间：2020-02-12T05:20:28Z

Image analysis tutorial 2 Scaling 2D splines so they measure in mm

简介：in this video we'll talk about scaling to real work coordinates um it's not absolutely necessary to do this but for your data to make sense you probably want all of the splines in the image to have an appropriate real-world scale

发布时间：2020-02-12T02:25:40Z

Ultrax: Vocal Tract Imaging for Speech Therapy and Research

简介：The Ultrax project (http://www.ultrax-speech.org) is a three-year project, funded by the EPSRC Healthcare Partnerships programme, which aims to use ultrasound to provide real-time visual feedback of lingual articulation. The ultimate aim is to exploit ultrasound to track tongue movement, while a simplified, diagrammatic vocal tract is displayed to the user as a realtime animation

发布时间：2020-02-07T03:30:46Z

A New Head-Probe Stabilization Device for synchronized Ultrasound and Electromagnetic Articulography recordings

简介：In this contribution,we will present and discuss a series of tests performed by means of a Toshiba Aplio XV ultrasound equipment and both the AG500 and the new AG501 electromagnetic articulographs by Carstens Medizinelektronik GmbH. Provided that the headset is absolutely transparent to the EMA operation, the aim of the tests is to show that the helmet and the ultrasound probe do not significantly move over the subject, resulting in reliable ultrasound tongue images......

发布时间：2020-02-07T03:08:41Z

Factors affecting the articulatory and acoustic characteristics of vowels in clear speech

简介：The purpose of this study was twofold: to examine the specific articulatory processes of clear speech for individual vowels using ultrasound, and to provide a more comprehensive investigation of the factors affecting the production of clear speech. In particular, we investigated how the type of listener (hard of hearing vs. normal hearing) and the lexical properties of words (frequency and phonological neighborhood density) interact.

发布时间：2020-02-07T02:47:43Z

An ultrasound study of oral and nasal central vowels in Brazilian Portuguese

简介：This s study aims at characterizing, by means of ultras sound techniques and acoustic analysis, the d differences in tongue position between the centraal oral and nasal vowels in syllable stressed d position and the central vowel in stressed and unstressed positions in Brazilian Portuguuese (BP). It is also concerned with approaching g its findings with the results of BP vowel desc criptions based on cine-radiographic data, acoustic analysis, MRI (Demolin & Medeiros, 2006; Gregioo,2006), and US.

发布时间：2020-02-07T01:47:31Z

The articulation of Scottish Gaelic plain and palatalized consonants

简介：Scottish Gaelic exhibits a rich system of consonant mutation, which is mostly governed by its morphology. For instance, the initial consonant [p] of a SG word bàta ‘boat’ changes to [v] when the word undergoes morphological inflection – e.g., a bhàta ‘his boat’, in which bh is pronounced as [v]. Although consonant mutation in SG is well documented (Macaulay, Ladefoged et al. , Gillies, Stewart ), there is only one experimental, quantitative study on this to date (Archangeli et al. (2011)).

发布时间：2020-02-06T10:30:10Z

An ultrasound study of Mandarin fricatives and affricate consonants

简介：This presentation will describe the articulatory contrasts between alveolar, post-alveolar, palatalized post-alveolar fricative and affricate consonants in Mandarin using the utrasound technique. The aim is to compare tongue contours between the three sets of consonants (alveolar, post-alveolar, palatal and alveo-palatal) and to describe how they are articulated. The present state of the study study is based on recordings made with 3 Mandarin speakers using the following set of data.

发布时间：2020-02-06T10:11:41Z

Contextual patterns of /l/-darkening in accents of English

简介：The phenomenon of /l/-darkening, whereby /l/ is produced with a delayed tongue-tip gesture, exhibits a remarkable amount of variation depending on its position in the word or phrase. Traditionally, it is said that light [l] occurs in onsets (e.g. light) and dark [l+] in codas (e.g. dull; Giegerich 1992; Halle & Mohanan 1985), but many studies report varying distributions of light and dark /l/ in different morphosyntactic environments.

发布时间：2020-02-06T10:01:20Z

An Exploratory Ultrasound Investigation of Emphatic Articulation in Cairene Arabic

简介：This study presents an exploratory quantitative investigation of emphatic articulation, employing Smoothing Spline Analysis of Variance (SS-ANOVA; Davidson, 2006) to statistically evaluate the differences in position between individual sections of the tongue during the articulation of plain and emphatic consonants.

发布时间：2020-02-06T09:49:29Z

Quantitative analysis of ultrasound data collected with and without head stabilisation

简介：The aim of the present study is to explore the applicability of quantitative measurements of stabilisation-free tongue movement data, by comparing ultrasound data collected with and without head stabilisation. The stabilisation device used in the study is a headset holding the transducer (Articulate Instruments Ltd 2008).

发布时间：2020-02-06T09:33:11Z

A Comparison of Methods for Tongue Shape Analysis

简介：The aim of the present study is to meaningfully describe through the use of a quantification parameter the differences in complexity between tongue shapes from utterances of various types, and to test whether a useful range for such a complexity parameter can be captured by differing candidate methods for a normal population. The most promising methods can then be used in further work to index differences in tongue shape complexity between normal and clinical populations.

发布时间：2020-02-06T09:20:20Z

Probe Skewing Effects & Kinect Head-tracking in Tongue Contour Imaging

简介：The position of the ultrasound probe in relation to the position of the head of the subject (the tilt of the head or probe) may affect the quality and comparability of ultrasound tongue images, compromising the interpretability of the data.

发布时间：2020-02-06T08:59:25Z

Segment Constants and Speaker Constants in English Coronal Obstruents

简介：This paper examines the articulation of the coronal obstruents /t, s, θ, tʃ, ʃ/. Previous studies have found that there is variation in the production of coronals (Dart, 1991; Delattre & Freeman, 1968; Reichard et al., 2012; Stone et al., 2012). These studies examined single segments (/r/, /s/). We compared segments with different places and manners of articulation, with the aim of isolating the articulatory correlates of the various coronal places in English.

发布时间：2020-02-06T08:41:30Z

Ultrasound images in the new ‘iPA Phonetics’ App

简介：iPA Phonetics is an application that illustrates the sounds and articulations of an expanded version of the IPA chart. The App gives users of Apple iOS mobile electronic devices the ability to access and compare and, through matching ‘games,’ to test their knowledge of phonetic symbols and sounds together with their visual production correlates, including video of the oral vocal tract, video of the laryngeal vocal tract, and ultrasound of the laryngeal vocal tract.

发布时间：2020-02-06T08:13:13Z

Statistical analysis of tongue shape in ultrasound video sequences: tongue tracking and population analysis

简介：In this study, we developed and evaluated a set of automated and semi-automated image processing tools for the statistical analysis of tongue shape, as seen in the midsagittal view on 2D ultrasound video sequences. It is known that tongue shape and motion, though highly non-rigid, are also highly constrained: most speech patterns can be explained by a few degrees of freedom.

发布时间：2020-02-06T07:33:56Z

Capturing, Analyzing, and Transmitting Intangible Cultural Heritage with the i-Treasures Project

简介：The i-Treasures project, which officially began on 1 February 2013, is a 12-partner FP7 project that proposes to use multi-sensor technology to capture, preserve, and transmit four types of intangible cultural heritage, referred to as ‘use cases’: rare traditional songs, rare dance interactions, traditional craftsmanship and contemporary music composition. Methodologies used will include body and gesture recognition, vocal tract modeling, speech processing and electroencephalography (EEG).

发布时间：2020-02-06T06:54:26Z

Can seeing help you hear? Acoustic vs. Ultrasound information in the diagnosis of speech sound disorders associated with cleft palate

简介：Many studies investigating speech characteristics in cleft palate (CP) have been based on perceptual analysis. In most cases, phonetic transcription is deemed to be gold standard in diagnosing speech sound disorders associated with cleft palate (Sell 2005). Due to the heterogeneity and complexity of CP, atypical and compensatory articulations may not be identified through phonetic transcription alone.

发布时间：2020-02-06T06:41:32Z

Retroflex versus bunched /r/ in ultrasound intervention: One shape does not fit all?

简介：Errors in speech sound production can have a negative impact on children’s development across academic, social, and psycho-emotional domains. While most children go on to attain age-appropriate speech production, an estimated 30% of English-speaking children with a history of speech sound disorder continue to misproduce rhotic sounds at 9 years of age, and 9% still show these errors from 1218 years and beyond (Lewis & Shriberg, 1994).

发布时间：2020-02-06T06:29:35Z

The Japanese unrounded back vowel /ɯ/ is in fact rounded central/front [ʉ - ʏ]

简介：This study reports on an ultrasound and video recording investigation of the so-called unrounded back vowel /!/ in Standard Japanese (SJ henceforth), as in /!ɕiɾo/ ‘back’, based on our hypothesis that /!/ is underlyingly a rounded central vowel with lip protrusion and the more appropriate phonemic IPA symbol is /ʉ/. As it turns out, the tongue position of SJ /!/ was closer to that of the front vowel /e/ than the back vowel /o/ by 6 of the 7 native SJ speaking participants.

发布时间：2020-02-06T06:14:18Z

/s/ in Italian-Tyrolean bilingual speakers

简介：This paper presents an exploratory UTI-based study on articulatory patterns for /s/C(C) clusters in Italian. More specifically, it focuses on word-initial and word-internal /s/-stop-(liquid) sequences as produced by four Italian/Tyrolean simultaneous and sequential bilingual speakers. The main aim of this research is to establish if the articulatory patterns for /s/ in Italian vary according to the nature and degree of bilingualism of the speaker.

发布时间：2020-02-06T04:11:22Z

What happens as listeners become speakers during face-to-face conversation?

简介：Approaching talk-in-interaction as an integrated system of body movement and vocalization, we collected simultaneous ultrasound and external video (Fig. 1) of eight participants’ face-to-face conversations with the first author. Focusing on inter-turn intervals as well as the period immediately following the start of a turn, we sought to ascertain if there were regular patterns among body movements as well as tongue postures,swallowing, and inhalation once speech had begun.

发布时间：2020-02-06T03:49:34Z

Ultrasound+corpus study of phonetically-motivated variation in North American French and English

简介：This presentation reports two projects employing ultrasound imaging to augment conversational speech corpora of French from Gatineau, Quebec, and English from Raleigh, North Carolina. Targeted articulatory data are collected in order to study phonetically-motivated sound changes in progress. While phonetically-motivated change from below is a fundamental concept in contemporary approaches to phonology and variation, empirical data is sparse (Cedergren, 1973; Trudgill, 1974; Labov, 2001)。

发布时间：2020-02-06T02:40:52Z

Kaytetye consonant contrasts without contours

简介：Two of the benefits of articulatory research using ultrasound technology are (a) the relative immediacy of access to visual information and (b) the portability of many ultrasound devices. Both points make ultrasound imaging an attractive option for field researchers interested in documenting lingual movement in remote locations.

发布时间：2020-02-06T02:16:23Z

Very High Frame Rate Ultrasound Tongue Imaging

简介：This paper examines the trade-off between temporal and spatial resolution in ultrasound tongue images at fast frame rates. The fastest lingual speech movements are investigated using a variety of echo pulse densities. Benefits and drawbacks of using higher frame rates are considered. Faster frame rates reduce distortion of the shape of the tongue during highly dynamic segments but it becomes increasingly difficult to discern the detail of that shape.

发布时间：2020-02-05T10:17:16Z

Predicting tongue shapes from a few landmark locations

简介：We present a method for predicting the midsagittal tongue contour from the locations of a few landmarks (metal pellets) on the tongue surface, as used in articulatory databases such as MOCHA and the Wisconsin XRDB. Our method learns a mapping using ground-truth tongue contours derived from ultrasound data and drastically improves over spline interpolation.

发布时间：2020-02-05T09:06:45Z

F1/F2 TARGETS FOR FINNISH SINGLE VS. DOUBLE VOWELS

简介：This paper explores the reason why Finnish single (short) vowels tend to occupy less peripheral positions in the F1/F2 vowel space compared to their double (long) counterparts. The results of two production studies suggest that the less extreme vowel quality of single vowels is best described as arising from undershoot of articulatory/acoustic targets due to their short durations, assuming single, context-free targets for phonemes.

发布时间：2020-02-05T07:41:36Z

AN ULTRASOUND PROTOCOL FOR COMPARING TONGUE CONTOURS: UPRIGHT VS SUPINE

简介：A study is described that employs ultrasound to measure the effects of gravity on production of vowels. The materials are designed to encourage consistent production over repetitions. A recording and analysis protocol is described which allows for correction for probe movement or rejection of data where correction is not possible. Results indicate a slight superior and posterior displacement of the tongue root in supine posture, consistent with a shift in the support structure of the tongue.

发布时间：2020-02-05T06:19:31Z

An Edinburgh Speech Production Facility

简介：The EPSRC‐funded Edinburgh Speech Production Facility is designed for the collection of synchronized articulatory and acoustic data during dialogue. It will be open to the international research community for funded use as of September 2010. As part of the EPSRC project, we recorded a corpus of dialogue sessions (8 are planned, 6 are recorded to date). The transcribed corpus will be available in 2010. We briefly describe the facility setup, and our recorded corpus.

发布时间：2020-02-05T05:38:43Z