Cross-corpora spoken language identification with domain diversification and generalization

Dey, Spandan; Sahidullah, Md; Saha, Goutam

doi:10.1016/j.csl.2023.101489

Cited by 8 publications

(1 citation statement)

References 95 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Results are expected to be independent of the specific characteristics of the corpus. Although with a different strategy to identify the domain, the DA training has been explored in other speech applications, such as automatic speech recognition [32], speech emotion recognition [33,34], spoken language identification [35], accent speech recognition [36], and voice conversion [37]. In the context of PD screening, a first attempt of applying domain adaptation was presented in [38], which used speaker identity-invariant representations from a single database (i.e., each domain is assumed to correspond to one speaker), but excluding multi-dataset scenarios.…”

Section: Introductionmentioning

confidence: 99%

Towards a Corpus (and Language)-Independent Screening of Parkinson’s Disease from Voice and Speech through Domain Adaptation

Ibarra,

Arias-Londoño,

Zañartu

et al. 2023

Bioengineering

View full text Add to dashboard Cite

End-to-end deep learning models have shown promising results for the automatic screening of Parkinson’s disease by voice and speech. However, these models often suffer degradation in their performance when applied to scenarios involving multiple corpora. In addition, they also show corpus-dependent clusterings. These facts indicate a lack of generalisation or the presence of certain shortcuts in the decision, and also suggest the need for developing new corpus-independent models. In this respect, this work explores the use of domain adversarial training as a viable strategy to develop models that retain their discriminative capacity to detect Parkinson’s disease across diverse datasets. The paper presents three deep learning architectures and their domain adversarial counterparts. The models were evaluated with sustained vowels and diadochokinetic recordings extracted from four corpora with different demographics, dialects or languages, and recording conditions. The results showed that the space distribution of the embedding features extracted by the domain adversarial networks exhibits a higher intra-class cohesion. This behaviour is supported by a decrease in the variability and inter-domain divergence computed within each class. The findings suggest that domain adversarial networks are able to learn the common characteristics present in Parkinsonian voice and speech, which are supposed to be corpus, and consequently, language independent. Overall, this effort provides evidence that domain adaptation techniques refine the existing end-to-end deep learning approaches for Parkinson’s disease detection from voice and speech, achieving more generalizable models.

show abstract