Multi-class Detection of Pathological Speech with Latent Features: How does it perform on unseen data?

TL;DR

This study explores multi-class classification of pathological speech using latent features from wav2vec 2.0, demonstrating robust performance on unseen data and various noise conditions.

Contribution

It introduces a multi-class approach leveraging latent features from wav2vec 2.0 for detecting multiple speech pathologies, extending beyond binary classification.

Findings

Achieves unweighted average F1-Scores between 74.1% and 97.0%.

Classifiers generalize well to unseen healthy speech data.

Robustness confirmed under noisy conditions with added room impulse responses.

Abstract

The detection of pathologies from speech features is usually defined as a binary classification task with one class representing a specific pathology and the other class representing healthy speech. In this work, we train neural networks, large margin classifiers, and tree boosting machines to distinguish between four pathologies: Parkinson's disease, laryngeal cancer, cleft lip and palate, and oral squamous cell carcinoma. We show that latent representations extracted at different layers of a pre-trained wav2vec 2.0 system can be effectively used to classify these types of pathological voices. We evaluate the robustness of our classifiers by adding room impulse responses to the test data and by applying them to unseen speech corpora. Our approach achieves unweighted average F1-Scores between 74.1% and 97.0%, depending on the model and the noise conditions used. The systems generalize and perform well on unseen data of healthy speakers sampled from a variety of different sources.