英国 icSpeech 便携式鼻音测量系统 Nasality Microphone

摘要 · 中文整理：鼻化度是鼻音亢进的重要临床生物标志物，传统上按经鼻发出的声能量与经口鼻发出的总声能量之比计算（eNasalance）。本文提出一种新方法，用以梅尔频率倒谱系数（MFCC）训练的卷积神经网络（CNN）来计算鼻化度（mfccNasalance）。研究从三方面评估 mfccNasalance 的准确度：1）训练与测试数据来自同一方言或不同方言；2）测试数据在动态性上存在差异（如快速产出的交替运动音节与短词）；3）采用多种 CNN 配置（卷积核形状以及是否使用 1×1 逐点卷积）。研究录制了来自不同方言健康说话人的双通道鼻流计语音：哥斯达黎加为「更鼻化」，西班牙与智利为「较少鼻化」。CNN 模型的输入为从 250 ms 滑动窗计算的 39 维 MFCC 向量序列。测试数据录制于西班牙，包含短词（非动态）、句子（动态）与交替运动音节（动态）。CNN 模型的准确度定义为该模型输出的 mfccNasalance 与人类专家感知鼻音评分之间的 Spearman 相关。在同方言条件下，无论 CNN 配置如何，mfccNasalance 都比 eNasalance 更准确；使用 1×1 卷积核可提升动态语料的准确度（p < .000），但对非动态语料无效；卷积核形状仅对非动态语料有显著影响（p < .000）。在跨方言条件下，得分显著低于同方言条件，尤以哥斯达黎加数据训练的模型为甚。我们的结论是：mfccNasalance 是 eNasalance 的一种灵活而有用的替代方案；未来研究应探索如何依据目标语音数据的动态性选择最合适的 CNN 模型以优化 mfccNasalance。

原文摘要：Nasalance is a valuable clinical biomarker for hypernasality. It is computed as the ratio of acoustic energy emitted through the nose to the total energy emitted through the mouth and nose (eNasalance). A new approach is proposed to compute nasalance using Convolutional Neural Networks (CNNs) trained with Mel-Frequency Cepstrum Coefficients (mfccNasalance). mfccNasalance is evaluated by examining its accuracy: 1) when the train and test data are from the same or from different dialects; 2) with test data that differs in dynamicity (e.g. rapidly produced diadochokinetic syllables versus short words); and 3) using multiple CNN configurations (i.e. kernel shape and use of 1 × 1 pointwise convolution). Dual-channel Nasometer speech data from healthy speakers from different dialects: Costa Rica, more(+) nasal, Spain and Chile, less(-) nasal, are recorded. The input to the CNN models were sequences of 39 MFCC vectors computed from 250 ms moving windows. The test data were recorded in Spain and included short words (-dynamic), sentences (+dynamic), and diadochokinetic syllables (+dynamic). The accuracy of a CNN model was defined as the Spearman correlation between the mfccNasalance for that model and the perceptual nasality scores of human experts. In the same-dialect condition, mfccNasalance was more accurate than eNasalance independently of the CNN configuration; using a 1 × 1 kernel resulted in increased accuracy for +dynamic utterances (p < .000), though not for -dynamic utterances. The kernel shape had a significant impact for -dynamic utterances (p < .000) exclusively. In the different-dialect condition, the scores were significantly less accurate than in the same-dialect condition, particularly for Costa Rica trained models. We conclude that mfccNasalance is a flexible and useful alternative to eNasalance. Future studies should explore how to optimize mfccNasalance by selecting the most adequate CNN model as a function of the dynamicity of the target speech data.

关键词：言语、语音信号处理、音节、卷积、元音、声学信号、语言、神经网络

研究目标：传统鼻音化率只依赖窄带能量比，既容易受说话人解剖与方言差异干扰，也难以刻画快速语流中的鼻音化动态，与听感判断的一致性有限；研究要解决的是能否用鼻音计双通道信号的倒谱特征结合卷积神经网络得到与专家听感更一致的鼻音化指标，并明确其方言迁移能力与网络配置选择依据。

主要结论：在训练与测试同方言的条件下，mfccNasalance 与专家听感鼻音评分的 Spearman 相关高于传统 eNasalance，且这一优势不依赖具体网络配置；1×1 逐点卷积显著提升高动态性语料（句子、交替轮替音节）的准确性，而卷积核形状只对低动态性的短词有显著影响；跨方言条件下准确性显著下降，尤其是以哥斯达黎加数据训练的模型。因此 mfccNasalance 是传统鼻音化率的有效替代，但需按目标语料的动态性与方言背景选择模型。

创新点：首次将鼻音计输出的双通道信号转换为 MFCC 序列并交由卷积神经网络学习，摆脱了固定窄带能量比的限制；利用方言本身鼻音程度的差异构造训练数据，避开对大规模病例标注的依赖；系统遍历 126 种卷积核组合并单独评估 1×1 逐点卷积，揭示了网络配置与语料动态性之间的匹配规律。

后续建议：纳入男性与更宽年龄段说话人，检验模型在性别与年龄上的泛化能力；构建跨方言自适应或迁移学习方案，缓解跨方言条件下准确性下降的问题；扩大临床样本并与腭咽功能的影像学或鼻咽镜结果对照，验证指标的临床效度；探索按语料动态性自动选择网络配置的自适应框架，减少人工调参；开展多中心、多品牌鼻音计的采集实验，验证方法对采集设备的稳健性

APA 引文：Lozano, A., Nava, E., García Méndez, M. D., & Moreno-Torres, I. (2024). Computing nasalance with MFCCs and Convolutional Neural Networks. PLOS ONE, 19(12), e0315452. https://doi.org/10.1371/journal.pone.0315452

仪器引用：The speech signal was recorded using a Nasometer (icSpeech, Rose Medical Solutions Ltd., Canterbury, UK), which produces two-channel audio signals.

摘要 · 中文整理：鼻音测量是一种无创、易操作的技术，可客观提供言语过程中口腔与鼻腔声能的数据。本研究的目标：一是编制西班牙语测试语料并确定 3 至 5 岁正常西班牙语儿童的鼻音化率常模值，使同龄的过高鼻音与腭咽功能障碍儿童有可比对的参照；二是判定鼻音化率分数是否存在显著的年龄或性别差异；三是检验元音负荷对鼻音化率的影响。研究采集了 130 名无已知言语、语言或听力障碍儿童的数据，按年龄分为三组。研究者在 MacKay-Kummer 简化鼻音测量评估程序修订版（SNAP-R，2005）基础上改编出西班牙语鼻音评估协议，要求全部受试者重复语音刺激。数据使用 icSpeech 鼻音计采集，并记录于 EXCEL 365 表格以供后续分析。由此获得了西班牙语版 SNAP-R 测验的常模数据。结果显示性别之间无显著差异，但年龄之间存在显著差异；同时发现高元音、前元音（尤其是音位 /i/）负荷会提高鼻音化率分数。该协议构建了与英文 SNAP-R 测验相对应的西班牙语语料，并为西班牙语儿童建立了常模。研究确认高元音会导致更高的鼻音化率，因而决定某段语料鼻音化率的是其元音构成，而非语料所属的语言。

原文摘要：Nasometry is a non-invasive, easy-to-use technique that provides objective data regarding the amount of acoustic energy in the oral and nasal cavities during speech. The goal of this study is to create a Spanish passage and determine normative nasalance values for typical Spanish-speaking children, 3 to 5 years of age, that allows us to compare the values in children of the same age who present hypernasality and velopharyngeal dysfunction. Second, to determine if there are significant differences in nasalance scores based on age or gender. Third, to test the impact of vowel load on nasalance scores. Data were collected from 130 children with no known speech, language, or hearing disorder. Participants were organized into three groups according to age. A nasometric assessment protocol for Spanish was developed based on an adaptation of the MacKay-Kummer Simplified Nasometric Assessment Procedures-Revised (SNAP-R; 2005). All participants were required to repeat the speech stimuli. Data were collected using the icSpeech nasometer and were recorded on EXCEL 365 sheets for further analysis. Normative data is now available for a Spanish version of the SNAP-R Test. No significant differences were found between the genders, but significant differences were found based on age. It was also found that a load of high and anterior vowels, especially the phoneme /i/, increased nasalance scores. This protocol created passages in Spanish that mimic the passages of the SNAP-R Test in English. These Spanish passages were normed for Spanish-speaking children. This study confirms that high vowels result in higher nasalance values and therefore, the vowel composition of a specific passage determines the nasalance score, not the language of the passage.

关键词：鼻音计、鼻音化率、共鸣、3至5岁儿童、西班牙语、元音负荷

研究目标：西班牙语儿童缺乏可用的现代鼻音化率常模数据，既有的西班牙语儿童常模来自 20 世纪 70 年代的 TONAR 鼻音计且已不再使用，导致临床上无法为 3 至 5 岁疑似过高鼻音、腭咽功能不全的患儿提供同龄对照；同时年龄、性别与元音负荷对鼻音化率的作用在既往研究中结论不一，需要系统检验。

主要结论：使用 Rose Medical Solutions 鼻音计采集的鼻音数据具有很高的可靠性（全协议 Cronbach's α 为 0.962，组内与组间重复测量均无显著差异且相关显著）；3 至 5 岁西班牙语儿童的鼻音化率不存在性别差异，但存在年龄差异，表现为 4 岁时下降、5 岁时回升并高于 3 岁；高元音、前元音尤其 /i/ 会明显抬高鼻音化率，因此决定一段语料鼻音化率的是其元音构成而非语言本身，研究据此给出了西班牙语版 SNAP-R 各刺激项的均值、标准差与正常阈值。

创新点：首次为 3 至 5 岁西班牙语儿童建立与英文 SNAP-R 测验结构对应的西班牙语鼻音评估语料及常模阈值；在 130 名儿童的大样本上验证了该鼻音计的内部一致性与重复测量信度；通过在同一协议内平行设置高元音与低元音刺激，直接分离出元音负荷而非语言差异对鼻音化率的贡献。

后续建议：扩大年龄范围，纳入 6 岁以上儿童及成人，考察鼻音化率随生长发育的完整变化轨迹；纳入腭裂、腭咽功能不全等临床组，检验所建阈值在真实筛查中的灵敏度与特异度；覆盖拉丁美洲等不同西班牙语方言区，验证常模的跨方言适用性；增加朗读段落与自发言语任务，以评估更长、更自然语料上的鼻音化率表现；将本仪器与其他品牌鼻音计做同步对照，给出跨设备的换算关系以便不同中心数据互通

APA 引文：Peris-Hernández, C., Rosell-Clari, V., & Kummer, A. W. (2024). Nasalance scores in Spanish-speaking children aged 3 to 5 years according to gender, age, and vowel load. Revista de Investigación en Logopedia, 14(1), e88054. https://doi.org/10.5209/rlog.88054

仪器引用：The Rose Medical Solutions, Ltd. (2019) nasometer, which is widely used in Europe, was used for data collection in this study. The specification data and system requirements for this nasometer are included in Appendix 1. The nasometer was used according to the manufacturer's instructions and calibration was performed prior to examination. … After the recording of each speech sample, a nasalance score was calculated using the Rose Medical Solutions nasometer software, iSpeech Professional Edition.