Glottal Source Information for Pathological Voice Detection

Abstract

Automatic methods for the detection of pathological voice from healthy speech can be considered as potential clinical tools for medical treatment. This study investigates the effectiveness of glottal source information in the detection of pathological voice by comparing the classical pipeline approach to the end-to-end approach. The traditional pipeline approach consists of a feature extractor and a separate classifier. In the former, two sets of glottal features (computed using the quasi-closed phase glottal inverse filtering method) are used together with the widely used openSMILE features. Using both the glottal and openSMILE features extracted from voice utterances and the corresponding healthy/pathology labels, support vector machine (SVM) classifiers are trained. In building end-to-end systems, both raw speech signals and raw glottal flow waveforms are used to train two deep learning architectures: (1) a combination of convolutional neural network (CNN) and multilayer perceptron (MLP), and (2) a combination of CNN and long short-term memory (LSTM) network. Experiments were carried out using three publicly available databases, including dysarthric (the UA-Speech database and the TORGO database) and dysphonic voices (the UPM database). The performance analysis of the detection system based on the traditional pipeline approach showed best results when the glottal features were combined with the baseline openSMILE features. The results of the end-to-end approach indicated higher accuracies (about 2-3 % improvement in all three databases) when glottal flow was used as the raw time-domain input (87.93 % for UA-Speech, 81.12 % for TORGO and 76.66 % for UPM) compared to using raw speech waveform (85.12 % for UA-Speech, 78.83 % for TORGO and 73.71 % for UPM). The evaluation of both approaches demonstrate that automatic detection of pathological voice from healthy speech benefits from using glottal source information. INDEX TERMS Pathological voice, glottal source waveform, glottal features, support vector machines, end-to-end systems.

Investigation and Evaluation of Glottal Flow Waveform for Voice Pathology Detection

“…Features used Classifier Database Best accuracy(%) [10] Autocorrelation and entropy SVM MEEI, SVD and AVPD 99.69, 92.79 and 99.79 [21] Glottal flow parameters SVM MEEI and SVD 99.27 and 93.66 [19] MFCC DBSCAN-SVM MEEI 98.63 [8] Jitter, Shimmer, HNR, TNI and NFHE KNN MEEI 96.1 [11] PPE SVM Private 91.4 [12] Largest Lyapunov exponent SVM MEEI 88.89 [22] openSMILE features and Glottal parameters SVM and CNN UA-speech and TORGO 87.93 and 76.66 [16] MFCC SVM SVD 86 [20] MFCC-QCP and Glottal source features SVM HUPA and SVD 78.37 [17] MLSF GMM MEEI 77.9 [9] HNR, NNE and GNE KNN UAM 66.57 which can represent the influence of voice diseases on the mechanism of the vocal folds.…”

“…Hence, the glottal flow waveform may have more potential for the detection and classification of pathological voices. For studies of glottal flow waveform, most previous works only evaluated the contribution of classical 12 glottal source parameters to voice pathology detection [20] [21] [22]. However, in this work, not only were classical glottal source parameters evaluated, but also the effects of acoustic features directly extracted from glottal source signals on pathological voice detection were explored.…”

“…The experimental results indicated that the glottal source features alone performed better than the spectral features; furthermore, the combination of all features used achieved the best accuracy of 78.37%. Recently, an investigation regarding the effectiveness of glottal source information in voice pathology detection was conducted by comparing the pipeline method to the end-to-end method in [22]. For the pipeline approach, glottal features computed using the QCP method and openSMILE features were used to train SVM.…”

A Classification Method for Judging the Depth of Chest Compression Based on CNN

“…In the model, the LSTM-based AE network extract ECG signal features, and the SVM classifier is applied for classifying different ECG arrhythmias signals [13].The result shows the proposed method has more than 99% accuracy. The author of [14] proposed a methods for the detection of pathological voice from healthy speech based on glottal source information. Two combination method are used, including a combination of convolutional neural network (CNN) and multilayer perceptron (MLP), and a combination of CNN and LSTM network.…”

Recognition of Abnormal Chest Compression Depth Using One-Dimensional Convolutional Neural Networks

“…The results show that the proposed method achieves more than 99% accuracy. The author of [ 14 ] proposed methods for detecting pathological voices from among healthy speech data, using glottal source information. Here, two combination of methods were used: a combination of a convolutional neural network (CNN) and multilayer perceptron, and a combination of a CNN and LSTM networks.…”