Automated Detection of the Competency of Delivering Guided Self-Help for Anxiety via Speech and Language Processing

Abstract

Speech and language play an essential role in automatically assessing several psychotherapeutic qualities. These automation procedures require translating the manual rating qualities to speech and language features that accurately capture the assessed psychotherapeutic quality. Speech features can be determined by analysing recordings of psychotherapeutic conversations (acoustics), while language-based analyses rely on the transcriptions of such psychotherapeutic conversations (linguistics). Guided self-help is a psychotherapeutic intervention that mainly relay on therapeutic competency of practitioners. This paper investigates the feasibility of automatically analysing guided self-help sessions for mild-to-moderate anxiety to detect and predict practitioner competence. This analysis is performed on sessions drawn from a patient preference randomised controlled trial using actual patient-practitioner conversations manually rated using a valid and reliable measure of competency. The results show the efficacy and potential of automatically detecting practitioners’ competence using a system based on acoustic and linguistic features extracted from transcripts generated by an automatic speech recogniser. Feature extraction, feature selection and classification or regression have been implemented as blocks of the prediction model. The Lasso regression model achieved the best prediction results with an R of 0.92 and lower error rates with an MAE of 1.66 and RMSE of 2.25.

Cognitive–behavioural versus cognitive–analytic guided self-help for mild-to-moderate anxiety: a pragmatic, randomised patient preference trial

“…Assessment of GSH competency has recently been automated using a system based on acoustic and linguistic features generated by an automatic speech recogniser to reduce time taken, as this has been acknowledged as a very time-consuming task for PWP clinical supervisors in routine practice. 25 Comparison of treatments for mild-to-moderate anxiety…”

Dementia Detection from Speech: What If Language Models Are Not the Answer?

“…In the context of the identification of mental health problems, the idea of using language models, in particular, has attracted a large amount of attention over the past few years. The automatic detection of depression [24][25][26][27], anxiety [28], bipolar disorder [29], and even schizophrenia [30] using advanced language models such as BERT [31], GPT-3, and GPT-4 [15] are some of the research directions in this context. However, the power of these language models comes at a cost.…”

Special Issue on Applications of Speech and Language Technologies in Healthcare

“…The four papers focused on detecting speech pathologies present innovative approaches. In [1], acoustic and linguistic features that are automatically extracted from patient-practitioner conversations are successfully used to predict practitioners' competency. Ozbolt et al [2] analysed methodological aspects in detecting speech pathology associated with Parkinson's disease from sustained vowels and provided recommendations to avoid misleading or overoptimistic results.…”