Comparing unsupervised speech learning directly to human performance in speech perception

Juliette Millet 1, 2 Nika Jurov 2 Ewan Dunbar 1, 2
1 CoML - Apprentissage machine et développement cognitif
LSCP - Laboratoire de sciences cognitives et psycholinguistique, Inria de Paris
Abstract : We compare the performance of humans (English and French listeners) versus an unsupervised speech model in a perception experiment (ABX discrimination task). Although the ABX task has been used for acoustic model evaluation in previous research, the results have not, until now, been compared directly with human behaviour in an experiment. We show that a standard, well-performing model (DPGMM) has better accuracy at predicting human responses than the acoustic baseline. The model also shows a native language effect, better resembling native listeners of the language on which it was trained. However, the native language effect shown by the models is different than the one shown by the human listeners, and, notably , the models do not show the same overall patterns of vowel confusions.
Complete list of metadatas

Cited literature [22 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02274499
Contributor : Ewan Dunbar <>
Submitted on : Thursday, August 29, 2019 - 9:14:32 PM
Last modification on : Monday, September 2, 2019 - 11:01:08 AM

File

Paper_Cogsci___rewrite-2.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02274499, version 1

Collections

Citation

Juliette Millet, Nika Jurov, Ewan Dunbar. Comparing unsupervised speech learning directly to human performance in speech perception. CogSci 2019 - 41st Annual Meeting of Cognitive Science Society, Jul 2019, Montréal, Canada. ⟨hal-02274499⟩

Share

Metrics

Record views

34

Files downloads

21