The effects of telepractice technology on dysarthric speech evaluation

Project Summary/Abstract With telepractice becoming an increasingly popular, indeed necessary, alternative to clinic visits, speech- language pathologists (SLPs) evaluate speech that is audio captured and transmitted remotely via a teleconferencing solutions (e.g. Zoom). All popular teleconferencing applications use speech compression

algorithms based on linear predictive coding (LPC) to reduce bandwidth required for speech transmission. LPC compression algorithms decompose speech into phonatory source and articulatory filter parameters, which are then independently vector quantized and temporally smoothed based on schemes that have been developed

and optimized for compressing speech sampled from the general population. In this way, speech is transmitted with the least number of audible artifacts and the highest level of intelligibility for the given internet connection constraints. Because LPC algorithms are optimized using large corpora of typical speech, they are

fundamentally not well-suited for faithfully transmitting the amount of distortion or noise commonly present in the dysarthric speech signal, and articulation and voice features are particularly vulnerable to corruption. In this proposal the aim is to systematically characterize the impact of speech compression algorithms commonly

used in telepractice platforms on speech intelligibility, perceptual evaluation, and acoustic measurement. This is done via two aims: SA1: Evaluate the effects of teleconferencing speech compression algorithms at three internet bandwidth levels on the perceptual and acoustic assessment of dysarthric speech

Existing high-fidelity audio recordings of words and sentences and sustained phonations from 20 speakers with various dysarthrias will be encoded at three compression rates to simulate low-, moderate-, and high- bandwidth internet connectivity. Twenty SLPs will transcribe the samples to attain intelligibility measures for the

original and encoded words and sentences; and perceptually rate vocal quality on sustained phonations. Acoustic measures of articulation and voice will be extracted. Within-subject statistical models will evaluate impact of bandwidth condition on perceptual and acoustic outcomes within speech tasks.

SA2: Compare outcomes for dysarthric speech recorded in a telepractice session (compressed) versus that recorded simultaneously in-person via smartphone application (uncompressed). Fifteen speakers with dysarthria will participate in simulated telepractice speech assessments administered by SLPs. Subsequently, recordings from session (compressed) and simultaneously recorded in-person samples

(uncompressed) will be scored by SLPs. In-person recordings will also be compressed as in SA1 conditions. Acoustic metrics will be extracted from all samples. Within-subject statistical models will evaluate sample differences across conditions (uncompressed, compressed during telepractice, and low- moderate- and high-

bandwidth compression levels). Results will inform the limits of telepractice for dysarthria evaluation.