/EmbodiedCNN-Speech

Embodied approach for Recognizing Spoken Digits from the Google Tensorflow Speech Commands datasets

Primary LanguagePythonGNU Lesser General Public License v3.0LGPL-3.0

An embodied model of Spoken Digits Recognition

Speech recognition - recognizing digits from Google Speech Commands dataset with embodied CNN architectures Python-Keras scripts for training and testing models. These are provided to facilitate replicating the results and experimenting other configurations and parameters.

The script ran in a nvidia-docker environment, Keras 2.2.4 and Tensorflow 1.8.

Folders:

Dataset - includes iCub robot finger representations for digits and scripts to create them from the Google Speech Commands dataset v0.02

Zipfian - scripts for the experiment with the Zipfian distribution (Scenario 1)

Longitudinal - scripts for the longitudinal experiment (Scenario 2)

Reference: Di Nuovo, A., McClelland, J.L. Developing the knowledge of number digits in a child-like robot. Nature Machine Intelligence 1, 594–605 (2019) doi:10.1038/s42256-019-0123-3

Free access link: https://rdcu.be/bYKKP