Influence of Utterance and Speaker Characteristics on the Classification of Children with Cleft Lip and Palate

TL;DR

This study demonstrates that features from pre-trained wav2vec 2.0 models can accurately classify children with Cleft Lip and Palate from healthy controls, but classifier reliability decreases with domain mismatches.

Contribution

It introduces the use of wav2vec 2.0 latent representations for CLP classification and analyzes factors affecting classifier robustness across different domains.

Findings

Latent representations from wav2vec 2.0 can achieve 100% accuracy in distinguishing CLP from healthy voices.

Classifier performance drops significantly with domain mismatches such as age, spoken content, and acoustic conditions.

Lower and middle encoder layers of wav2vec 2.0 are most effective for classification.

Abstract

Recent findings show that pre-trained wav2vec 2.0 models are reliable feature extractors for various speaker characteristics classification tasks. We show that latent representations extracted at different layers of a pre-trained wav2vec 2.0 system can be used as features for binary classification to distinguish between children with Cleft Lip and Palate (CLP) and a healthy control group. The results indicate that the distinction between CLP and healthy voices, especially with latent representations from the lower and middle encoder layers, reaches an accuracy of 100%. We test the classifier to find influencing factors for classification using unseen out-of-domain healthy and pathologic corpora with varying characteristics: age, spoken content, and acoustic conditions. Cross-pathology and cross-healthy tests reveal that the trained classifiers are unreliable if there is a mismatch between training and out-of-domain test data in, e.g., age, spoken content, or acoustic conditions.