Pre-trained weights for the baseline DNN system of DCASE 2021 automated audio captioning task

This is the repository of the pre-trained weights for the baseline deep neural network (DNN), used in the baseline system of automated audio captioning at the DCASE 2021 Challenge. 

The pre-trained weights can be used with the baseline DNN in order to reproduce the reported results on the evaluation split (development-testing set in DCASE terminology) of the Clotho dataset. 

You can find the description of the automated audio captioning task and the reported results on the webpage of the task: http://dcase.community/challenge2021/task-automatic-audio-captioning  

Clotho dataset can be found at: https://zenodo.org/record/3490684

GitHub repositories of audio captioning can be found at: https://github.com/audio-captioning

If you use the baseline system, please consider citing the paper of Clotho: 

K. Drossos, S. Lipping, and T. Virtanen, "Clotho: An Audio Captioning Dataset," to be presented in the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), May 4-8, 2020

available online at: https://arxiv.org/abs/1910.09387