Evaluating automatic speech recognition systems as quantitative models of cross-lingual phonetic category perception

Thomas Schatz (1) , Francis Bach (2, 3) , Emmanuel Dupoux (1, 4)

1 LSCP - Laboratoire de sciences cognitives et psycholinguistique
2 LIENS - Laboratoire d'informatique de l'école normale supérieure
3 SIERRA - Statistical Machine Learning and Parsimony
4 CoML - Apprentissage machine et développement cognitif

Theories of cross-linguistic phonetic category perception posit that listeners perceive foreign sounds by mapping them onto their native phonetic categories, but, until now, no way to effectively implement this mapping has been proposed. In this paper, Automatic Speech Recognition systems trained on continuous speech corpora are used to provide a fully specified mapping between foreign sounds and native categories. The authors show how the machine ABX evaluation method can be used to compare predictions from the resulting quantitative models with empirically attested effects in human cross-linguistic phonetic category perception.

Domaines: Linguistique; Sciences cognitives

Thomas Schatz, Francis Bach, Emmanuel Dupoux. Evaluating automatic speech recognition systems as quantitative models of cross-lingual phonetic category perception. Journal of the Acoustical Society of America, 2018, 143 (5), pp.EL372 - EL378. ⟨10.1121/1.5037615⟩. ⟨hal-01888735⟩