Automatic Voice Disorder Detection Using Self-Supervised Representations

Abstract

Many speech features and models, including Deep Neural Networks (DNN), are used for classification tasks between healthy and pathological speech with the Saarbruecken Voice Database (SVD). However, accuracy values of 80.71% for phrases or 82.8% for vowels /aiu/ are the highest reported for audio samples in SVD when the evaluation includes the wide amount of pathologies in the database, instead of a selection of some pathologies. This paper targets this top performance in the state-of-the-art Automatic Voice Disorder Detection (AVDD) systems. In the framework of a DNN-based AVDD system we study the capability of Self-Supervised (SS) representation learning for describing discriminative cues between healthy and pathological speech. The system processes the SS temporal sequence of features with a single feed-forward layer and Class-Token (CT) Transformer for obtaining the classification between healthy and pathological speech. Furthermore, there is evaluated a suitable data extension of the training set with out-ofdomain data is also evaluated to deal with the low availability of data for using DNN-based models in voice pathology detection. Experimental results using audio samples corresponding to phrases in the SVD dataset, including all pathologies available, show classification accuracy values until 93.36%. This means that the proposed AVDD system achieved accuracy improvements of 4.1% without the training data extension, and 15.62% after the training data extension compared to the baseline system. Beyond the novelty of using SS representations for AVDD, the fact of obtaining accuracies over 90% in these conditions and using the whole set of pathologies in the SVD is a milestone for voice disorder-related research. Furthermore, the study on the amount of in-domain data in the training set related to the system performance show guidance for the data preparation stage. Lessons learned in this work suggest guidelines for taking advantage of DNN, to boost the performance in developing automatic systems for diagnosis, treatment, and monitoring of voice pathologies.

Automatic Speech and Voice Disorder Detection Using Deep Learning—A Systematic Literature Review

“…The deep learning models commonly used in these various studies are summarized in Table 6. Ribas et al [1] proposed a very promising approach of using a self supervised(SS) representations for classifying healthy and pathological voices. SS representations were basically learned from unlabeled speech data, and then used as an input to a classification model that is trained on labeled speech data.The authors explored two modalities of SS representations: frozen and fine tuned.…”

“…Overall, eight disorders were detected in these research articles including four speech and four voice disorders as shown in table 4. Most of the studies did not focus on one specific disorder, however they covered various disorders present within the category of either speech or sound disorder for instance: [1] covered 4 speech disorders, [9] covered 3 speech disorders, [11] covered 4 voice disorders, [15] worked on 3 voice disorders and [16] focuses on 6 voice disorders. Few studies including [4] and [8] were too general and their focus was to classify just pathological voice and healthy voice.…”

“…The performance of a speech or voice disorder detection model can be measured using various performance met- Multi-layer Perceptron [1], [8] Deep Neural Network (DNN) [2], [11], [15], [13] Convolutional Neural Network (CNN)…”

“…[3], [2], [5], [10], [9], [14], [16], [19], [20], [7] Long Short Term Memory (LSTM) [4], [3], [12], [2] Gated Recurrent Unit (GRU) [5], [3] Pyramid Bidirectional LSTM (PBLSTM) [13] Adaptive Optimization Artificial based Neural Network (AOANN) [21] Class Token(CT) Transformer [1] rics. These metrics include F-measure,accuracy, precision, recall, micro and macro averaging of accuracy, sensitivity, Unweighted Average and area under the curve.…”

Convolutional Neural Networks for Classification of Voice Qualities from Speech and Neck Surface Accelerometer Signals

“…The model learns representations from the raw speech, and these representations can be used in the required downstream task. Examples of pre-trained models are wav2vec2 and HuBERT that have shown good performance in various speech technology tasks, such as ASR, emotion recognition, speaker and language identification, and voice disorder detection [47,48,49,50,51,52,53]. There are, however, no studies on using recent self-supervised pre-trained models, such as wav2vec2 [47] and HuBERT [54], for voice quality classification.…”