Sarel van Vuuren

Relevance of time-frequency features for phonetic and speaker-channel classification

Abstract The mutual information concept is used to study the distribution of speech information in frequency and in time. The main focus is on the information that is relevant for phonetic classification. A large database of hand-labeled fluent speech is used to (a) compute the mutual information (MI) between a phonetic classification variable and one spectral feature variable in the time–frequency plane, and (b) compute the joint mutual information (JMI) between the phonetic classification variable and two feature variables in the time–frequency plane. The MI and the JMI of the feature variables are used as relevance measures to select inputs for phonetic classifiers. Multi-layer perceptron (MLP) classifiers with one or two inputs are trained to recognize phonemes to examine the effectiveness of the input selection method based on the MI and the JMI. To analyze the non-linguistic sources of variability, we use speaker-channel labels to represent different speakers and different telephone channels and estimate the MI between the speaker-channel variable and one or two feature variables.

Telerehabilitation, virtual therapists, and acquired neurologic speech and language disorders.

Telerehabilitation (telerehab) offers cost-effective services that potentially can improve access to care for those with acquired neurologic communication disorders. However, regulatory issues including licensure, reimbursement, and threats to privacy and confidentiality hinder the routine implementation of telerehab services into the clinical setting. Despite these barriers, rapid technological advances and a growing body of research regarding the use of telerehab applications support its use. This article reviews the evidence related to acquired neurologic speech and language disorders in adults, focusing on studies that have been published since 2000. Research studies have used telerehab systems to assess and treat disorders including dysarthria, apraxia of speech, aphasia, and mild Alzheimer disease. They show that telerehab is a valid and reliable vehicle for delivering speech and language services. The studies represent a progression of technological advances in computing, Internet, and mobile technologies. They range on a continuum from working synchronously (in real-time) with a speech-language pathologist to working asynchronously (offline) with a stand-in virtual therapist. One such system that uses a virtual therapist for the treatment of aphasia, the Web-ORLA™ (Rehabilitation Institute of Chicago, Chicago, IL) system, is described in detail. Future directions for the advancement of telerehab for clinical practice are discussed.

Advances in children's speech recognition within an interactive literacy tutor

In this paper we present recent advances in acoustic and language modeling that improve recognition performance when children read out loud within digital books. First we extend previous work by incorporating cross-utterance word history information and dynamic n-gram language modeling. By additionally incorporating Vocal Tract Length Normalization (VTLN), Speaker-Adaptive Training (SAT) and iterative unsupervised structural maximum a posteriori linear regression (SMAPLR) adaptation we demonstrate a 54% reduction in word error rate. Next, we show how data from childrens read-aloud sessions can be utilized to improve accuracy in a spontaneous story summarization task. An error reduction of 15% over previous published results is shown. Finally we describe a novel real-time implementation of our research system that incorporates time-adaptive acoustic and language modeling.

Acquisition and maintenance of scripts in aphasia: a comparison of two cuing conditions.

PURPOSE This study was designed to compare acquisition and maintenance of scripts under two conditions: high cue, which provided numerous multimodality cues designed to minimize errors, and low cue, which provided minimal cues. METHOD In a randomized controlled crossover study, eight individuals with chronic aphasia received intensive computer-based script training under two cuing conditions. Each condition lasted 3 weeks, with a 3-week washout period. Trained and untrained scripts were probed for accuracy and rate at baseline, during treatment, immediately posttreatment, and at 3 and 6 weeks posttreatment. Significance testing was conducted on gain scores, and effect sizes were calculated. RESULTS Training resulted in significant gains in script acquisition with maintenance of skills at 3 and 6 weeks posttreatment. Differences between cuing conditions were not significant. When severity of aphasia was considered, there also were no significant differences between conditions, although magnitude of change was greater in the high-cue condition versus the low-cue condition for those with more severe aphasia. CONCLUSIONS Both cuing conditions were effective in acquisition and maintenance of scripts. The high-cue condition may be advantageous for those with more severe aphasia. Findings support the clinical use of script training and the importance of considering aphasia severity.

A Virtual Therapist for Speech and Language Therapy

A virtual therapist (VT) capable of modeling visible speech and directing speech and language therapy is presented. Three perspectives of practical and clinical use are described. The first is a description of treatment and typical roles that the VT performs in directing participation, practice and performance. The second is a description of techniques for modeling visible speech and implementing tele-rehabilitation. The third is an analysis of performance of a system (AphasiaRx™) for delivering speech and language therapy to people with aphasia, with results presented from a randomized controlled cross-over study in which the VT provided two levels of cuing. Compared to low cue treatment, high cue treatment resulted in 2.3 times faster learning. The paper concludes with a discussion of the benefits of speech and language therapy delivered by the VT.

Impact of Personal Relevance on Acquisition and Generalization of Script Training for Aphasia: A Preliminary Analysis

PURPOSE The importance of personalization in script training in aphasia has been assumed but never tested. This study compared acquisition and generalization of personally relevant versus generic words or phrases appearing in the same scripts. METHOD Eight individuals (6 men; 2 women) with chronic aphasia received 3 weeks of intensive computer-based script training. For each participant, 2 scripts, a trained and an untrained generalization script, were embedded with 4 personally relevant word choices and 2-4 generic items that were similar across participants. Scripts were probed for accuracy at baseline and posttreatment. Significance testing was conducted on baseline and posttreatment scores, and for gains in personally relevant versus generic items. Effect sizes were computed. RESULTS Both personally relevant and generic items improved significantly on trained scripts. Improvements on untrained scripts were smaller, with only personally relevant items reaching significance. There was no significant difference on gains made on personalized versus generic items for trained scripts (p = .059), but the effect size was large (d = 0.90). For generalization scripts, this effect was small (d = 0.25) and nonsignificant. CONCLUSIONS Personally relevant words and phrases were acquired, although not generalized, more successfully than generic words and phrases. Data supports the importance of personalization in script training, but the degree of that importance requires further investigation.

Evaluating the impact of practice conditions (randomized vs. blocked) and schedule (distributed vs. massed) on script training in aphasia

Background: A growing body of evidence suggests that intensity of therapy is a critical factor influencing outcomes in aphasia treatment regardless of the treatment approach. However, little information is available regarding scheduling or practice conditions that optimize treatment outcomes. Much of the literature on practice and schedule conditions has come from well-established principles of motor learning. The extent to which these principles apply to aphasia, and to tasks that include not just motor skill learning but also complex language learning, requires investigation. Aims: The purpose of this study was to assess the impact of practice conditions (random vs. blocked) and schedule (distributed vs. massed) on script training in aphasia. Methods & Procedures: Thirty-six participants with aphasia (22 males; 14 females; 24 nonfluent; 12 fluent) were randomly assigned to one of four conditions that varied by practice (blocked vs. random) and schedule (massed vs. distributed). Overall, mean age was 58.57 (SD = 16.83) years; mean time post onset was 35.8 (SD = 35.9) months; mean education was 15.8 (SD = 2.6) years; and mean Western Aphasia Battery-Revised Aphasia Quotient was 62.96 (SD = 14.20). The treatment used was script training, administered via a computer program called AphasiaScripts that has demonstrated positive outcomes across several studies (Cherney, Halper, & Kaye, 2011; Cherney, Halper, Holland, & Cole, 2008; Cherney, Kaye, & Van Vuuren, 2014). To ensure consistency across subjects and across arms, standard scripts were used. The complexity of the scripts was set relative to the participant’s language level using methods successfully developed and previously reported (Kaye & Cherney, 2016). All participants received 18 h of script training on 12 different days (i.e. 3 × 30-min treatment sessions for a total of 90 min on each designated treatment day). Those randomized to the massed schedule arm completed the 18 h over a 2-week period; those randomized to the distributed schedule arm completed the 18 h over a 4-week period. In the blocked practice arm, participants practiced one script at a time. They practiced the first script for 9 h (6 treatment days) over 1 week in the massed condition, and over 2 weeks in the distributed condition. Then they practiced