PAoS Markers: Trajectory Analysis of Selective Phonological Posteriors for Assessment of Progressive Apraxia of Speech

Progressive apraxia of Speech (PAoS) is a progressive motor speech disorder associated with neurodegenerative disease causing impairment of phonetic encoding and motor speech planning. Clinical observation and acoustic studies show that duration analysis provides reliable cues for diagnosis of the disease progression and severity of articulatory disruption. The goal of this paper is to develop computational methods for objective evaluation of duration and trajectory of speech articulation. We use phonological posteriors as speech features. Phonological posteriors consist of probabilities of phonological classes estimated for every short segment of the speech signal. PAoS encompasses lengthening of duration which is more pronounced in vowels [1, 2]; we thus hypothesize that a small subset of phonological classes provide stronger evidence for duration and trajectory analysis. These classes are determined through analysis of linear prediction coefficients (LPC). To enable trajectory analysis without phonetic alignment, we exploit phonological structures defined through quantization of phonological posteriors. Duration and trajectory analysis are conducted on blocks of multiple consecutive segments possessing similar phonological structures. Moreover, unique phonological structures are identified for every severity condition.