Automatic Discrimination of Apraxia of Speech and Dysarthria Using a Minimalistic Set of Handcrafted Features

Abstract

To assist clinicians in the differential diagnosis and treatment of motor speech disorders, it is imperative to establish objective tools which can reliably characterize different subtypes of disorders such as apraxia of speech (AoS) and dysarthria. Objective tools in the context of speech disorders typically rely on thousands of acoustic features, which raises the risk of difficulties in the interpretation of the underlying mechanisms, overadaptation to training data, and weak generalization capabilities to test data. Seeking to use a small number of acoustic features and motivated by the clinical-perceptual signs used for the differential diagnosis of AoS and dysarthria, we propose to characterize differences between AoS and dysarthria using only six handcrafted acoustic features, with three features reflecting segmental distortions, two features reflecting loudness and hypernasality, and one feature reflecting syllabification. These three different sets of features are used to separately train three classifiers. At test time, the decisions of the three classifiers are combined through a simple majority voting scheme. Preliminary results show that the proposed approach achieves a discrimination accuracy of 90%, outperforming using state-of-the-art features such as openSMILE which yield a discrimination accuracy of 65%.

Automatic And Perceptual Discrimination Between Dysarthria, Apraxia of Speech, and Neurotypical Speech

“…The acoustic features proposed in this paper for discriminating between dysarthria, AoS, and neurotypical speech are motivated by the advantageous performance these features have shown in discriminating between dysarthria and neurotypical speech in [16] and in discriminating between dysarthria and AoS in [17]. In the following, a brief overview of these features is presented.…”

“…In [16], we have shown that spectral sparsity can successfully characterize imprecise articulation, abnormal pauses, and breathiness observed in dysarthria. Spectral sparsity describes the energy distribution of the speech spectral coefficients across time and is computed by i) transforming the signals to the short-time Fourier transform domain, ii) time-aligning all representations to a 1 For additional details on the motivation behind these features and their computation, the interested reader is referred to [16,17]. reference representation, and iii) computing the shape parameter of a Chi distribution best modeling the spectral magnitudes in each time frame [16].…”

“…Formant frequencies and duration of continuously voiced regions. As in [17], vowel distortion and inappropriate vowel lengthening commonly observed in AoS are characterized by the 10-dimensional feature vector f2 constructed from the statistics (i.e., mean, standard deviation, kurtosis, and skewness) of the first and second formant frequencies across time and the statistics (i.e., mean and standard deviation) of the duration of continuously voiced regions.…”

“…Typical automatic techniques operate on acoustic features which are handcrafted to reflect impaired speech dimensions. Many acoustic features have been successfully exploited to characterize impacted phonation and articulation, e.g., fundamental and formant frequencies, jitter, shimmer, Mel frequency cepstral coefficients, or temporal and spectral sparsity [11][12][13][14][15][16][17]. In an attempt to capture many impaired speech dimensions, also large-scale feature sets such as openSMILE have been used [18][19][20].…”

“…The majority of state-of-the-art contributions deal with impaired speech arising due to dysarthrias or laryngeal disorders, with AoS being considered only in [17]. These contributions propose twoclass techniques aiming to discriminate between dysarthria and neurotypical speech, laryngeal disorders and neurotypical speech, or dysarthria and AoS.…”