Please join us for an interdisciplinary seminar cycle about Benchmarking Speech Acquisition in Humans and Machines.

Recent advances in machine learning make unsupervised learning of speech feasible. But bridging the

gap between psycholinguistics and unsupervised machine learning will take theoretical and empirical

progress. This is particularly obvious in cases where naturalistic audio that faithfully represents infants'

experiences is used for training. Additionally, true benchmarking involves trying to directly relate

machines to humans, which is far from trivial.

The goal of the present cycle is to give an overview of the issues in benchmarking humans and

machines in terms of their acquisition of speech perception for members of both communities and

engage in discussions, as well as potentially empirical studies, to resolve open issues. The focus will be

put exclusively on work that attempts to approach speech perception in its full complexity, meaning

using data sets as realistic as possible (for humans and/or machines).

Kick-off meeting: March 8, time 4-6:15pm CET

Following meetings: 2nd Tue of each month, 4-5pm CET

The description of the seminar can be found here.

You can subscribe to the mailing list here.

Feel free to forward this email to anyone who might be interested.

Best,

--------------------------------------------------------------------------------------------

Alejandrina Cristia, Marvin Lavechin

Laboratoire de Sciences Cognitives et Psycholinguistique
29, rue d'Ulm, 75005, Paris, FRANCE

--------------------------------------------------------------------------------------------

BSA Benchmarking Speech Acquisition

Description