What question did this study set out to answer?

The aim is to develop automated, deep learning-based tools for analyzing pathological speech, improving clinical assessment and therapy.

February 2, 2026Open Access

Deep Learning Approaches for Clinical Speech Analysis: From Latent Representations to Articulatory Estimation

View Full Paper

Ask AI

Bookmark

Authors

Novel deep learning techniques enhance speech analysis in individuals recovering from stroke, suggesting innovative therapeutic tools.

The aim is to develop automated, deep learning-based tools for analyzing pathological speech, improving clinical assessment and therapy.
Developed a method for automatic intelligibility assessment using latent speech representations.
Investigated pre-training of self-supervised learning models on healthy speech versus pathological speech.
Created a multi-task learning architecture for estimating articulatory trajectories and phoneme sequences from raw audio.
Proposed a comprehensive framework for speech articulation analysis, integrating several analytic tasks.
Achieved high performance in predicting intelligibility scores with reduced reference signal reliance.
Demonstrated that pre-training exclusively on healthy speech avoids harmful biases.
Successfully predicted articulatory movements and phonetic information without relying on transcriptions.
The integrated framework shows promise for therapy-oriented applications in communication disorders.

Tobias Weise (2026) studied this question.

View Full Paper

Ask AI

Bookmark