Gavidia-Ceballos L, Hansen J H
Department of Biomedical Engineering, Duke University, Durham, NC 37708-0291, USA.
IEEE Trans Biomed Eng. 1996 Apr;43(4):373-83. doi: 10.1109/10.486257.
The focus of this study is to formulate a speech parameter estimation algorithm for analysis/detection of vocal fold pathology. The speech processing algorithm proposed estimates features necessary to formulate a stochastic model to characterize healthy and pathology conditions from speech recordings. The general idea is to separate speech components under healthy and assumed pathology conditions. This problem is addressed using an iterative maximum-likelihood (ML) estimation procedure, based on the estimation-maximization (EM) algorithm. A new feature for characterizing pathology, termed enhanced-spectral-pathology component (ESPC), is estimated and shown to vary consistently between healthy and pathology conditions. It is also shown that the mean-area-peak-value (MAPV) and the weighted-slope (WSLOPE) indexes, which are obtained from the ESPC estimate, are meaningful measures of speech pathology conditions. For classification purposes, a five-state hidden-Markov-model (HMM) recognizer was formulated, based on the MAPV, WSLOPE, and ESPC spectral features. A set of log Mel-frequency filter bank coefficients were used to parameterize the ESPC feature. An evaluation of the HMM-based classifier was performed using speech recordings from healthy and vocal fold cancer patients of sustained vowel sounds. It is shown that while both MAPV and WSLOPE are useful features for vocal fold pathology detection, superior performance was achieved using a finer spectral representation of ESPC (e.g., a detection rate of 88.7% for pathology and 92.8% for healthy condition). One main advantage of the proposed method is that it does not require direct estimation of the glottal flow waveform. Therefore, the limitation of the inability to characterize vocal fold pathology, due to incomplete glottal closure, is no longer an issue. The results suggest that general analysis of the ESPC feature can provide a quantitative, noninvasive approach for analysis, detection, and characterization of speech production under vocal fold pathology.
本研究的重点是制定一种语音参数估计算法，用于分析/检测声带病变。所提出的语音处理算法估计了构建随机模型所需的特征，以便根据语音记录来表征健康和病变状况。总体思路是在健康和假定的病变状况下分离语音成分。基于期望最大化（EM）算法，使用迭代最大似然（ML）估计程序来解决这个问题。估计了一种用于表征病变的新特征，称为增强频谱病变成分（ESPC），并表明其在健康和病变状况之间存在一致变化。还表明，从ESPC估计中获得的平均面积峰值（MAPV）和加权斜率（WSLOPE）指标是语音病变状况的有意义度量。为了进行分类，基于MAPV、WSLOPE和ESPC频谱特征构建了一个五状态隐马尔可夫模型（HMM）识别器。使用一组对数梅尔频率滤波器组系数对ESPC特征进行参数化。使用来自健康人和声带癌患者的持续元音语音记录对基于HMM的分类器进行了评估。结果表明，虽然MAPV和WSLOPE都是用于声带病变检测的有用特征，但使用ESPC更精细的频谱表示可实现更好的性能（例如，病变检测率为88.7%，健康状况检测率为92.8%）。所提出方法的一个主要优点是它不需要直接估计声门流波形。因此，由于声门闭合不完全而无法表征声带病变的局限性不再是一个问题。结果表明，对ESPC特征的一般分析可以为声带病变情况下语音产生的分析、检测和表征提供一种定量、非侵入性的方法。
