Towards an ASR-free objective analysis of pathological speech

Abstract

Nowadays, intelligibility is a popular measure of the severity of the articulatory deficiencies of a pathological speaker. Usually, this measure is obtained by means of a perceptual test, consisting of nonconventional and/or nonconnected words. In previous work, we developed a system incorporating two Automatic Speech Recognizers (ASR) that could fairly accurately estimate phoneme intelligibility (PI). In the present paper, we propose a novel method that aims to assess the running speech intelligibility (RSI) as a more relevant indicator of the communication efficiency of a speaker in a natural setting. The proposed method computes a phonological characterization of the speaker by means of a statistical analysis of frame-level phonological features. Important is that this analysis requires no knowledge of what the speaker was supposed to say. The new characterization is demonstrated to predict PI and to provide valuable information about the nature and severity of the pathology.

Combining phonological and acoustic ASR-free features for pathological speech intelligibility assessment

“…Based on these features, the approaches in [3] and [4] are applied to create two ASR-free speaker feature sets: an acoustical and a phonological feature set.…”

“…For the training and validation of our models we adopted a leave-one-out cross validation scheme. We tried two statistical learners for every IPM: one based on ensemble linear regression (ELR) with feature selection [4] and one based on Support Vector Regression (SVR) [12].…”

“…Another ASR-free approach was presented in [4]. It relies on a statistical analysis of the feature patterns emerging from phonological feature detectors which are trained on normal speech.…”

“…In this paper, we investigate whether the feature sets are really language-independent and whether combining them leads to a model that outperforms the individual models. For this purpose, we conduct experiments on the two datasets that were formerly used to test the individual approaches that were presented in [3] and [4] respectively.…”

Pathological voice adaptation with autoencoder-based voice conversion

“…However, synthesising pathological speech via VC is not without challenges. Existing pathological speech corpora [8,9,5,10] provide healthy control speakers, but healthy speech recordings from the same pathological speaker are rarely available. This means that a successful pathological voice conversion system needs to learn conversion of both, the voice and pathological characteristics simultaneously, as suggested in previous work [4].…”