Automatic Prediction of Speech Evaluation Metrics for Dysarthric Speech

Abstract

During the last decades, automatic speech processing systems witnessed an important progress and achieved remarkable reliability. As a result, such technologies have been exploited in new areas and applications including medical practice. In disordered speech evaluation context, perceptual evaluation is still the most common method used in clinical practice for the diagnosing and the following of the condition progression of patients despite its well documented limits (such as subjectivity). In this paper, we propose an automatic approach for the prediction of dysarthric speech evaluation metrics (intelligibility, severity, articulation impairment) based on the representation of the speech acoustics in the total variability subspace based on the i-vectors paradigm. The proposed approach, evaluated on 129 French dysarthric speakers from the DesPhoAPady and VML databases, is proven to be efficient for the modeling of patient's production and capable of detecting the evolution of speech quality. Also, low RMSE and high correlation measures are obtained between automatically predicted metrics and perceptual evaluations.

Automatic Evaluation of Speech Intelligibility Based on I-vectors in the Context of Head and Neck Cancers

“…Indeed, r and RMSE measures reach 0.84 and 2.339 respectively. This correlation rate is consistent with previous results observed over read speech produced by dysarthric patients [15]. Also, the resulting RMSE measure is quite low considering that the interval of the reference measure is characterized by a wide range ([0,22] for this corpus) and an extreme sensibility.…”

“…In previous work [15], we proposed an approach based on the i-vector paradigm for the automatic prediction of several dysarthric speech evaluation metrics like intelligibility, severity, and articulation impairment. The proposed approach was applied on 129 dysarthric and healthy speakers and high correlation measures (between 0.8 and 0.9) were reached between the different automatic predictions and reference perceptual speech evaluation metrics.…”

Automatic Assessment of Prosodic Quality in Down Syndrome: Analysis of the Impact of Speaker Heterogeneity

“…Some works focus on the speech intelligibility of people with aphasia [23,24] or speech intelligibility in pathological voices [25,26]. Others try to identify speech disorders in children with cleft lip and palate [27] or to predict automatically some dysarthric speech evaluation metrics, such as intelligibility, severity and articulation impairment [28,29]. In addition, the recognition of speech emotions and autism spectrum disorders has also been investigated [30].…”

Perceptual evaluation for automatic anomaly detection in disordered speech: Focus on ambiguous cases

“…To cope with these limitations, automatic approaches have rapidly emerged, as potential solutions, by providing objective tools for intelligibility assessment and anomaly detection in pathological speech [14]. In the literature, we distinguish two main approaches: those directly based on automatic speech transcription and word transcription error rate to compute an intelligibility score [15,16], and those using automatic speech processing techniques as a means of extracting relevant information from the speech signal to perform an automatic evaluation of speech on different granularities [17,18,19,20,21].…”