Can you please describe how the **clotho-dataset** directory look like?

[jiminbot20]

Can you please describe how the clotho-dataset directory look like?

is this capture below right?

And in your code, data split is not using?

[topel]

In ../clotho-dataset/data, there are these files:

characters_frequencies.p
words_list.p
words_frequencies.p
characters_list.p         

clotho_captions_evaluation.csv   
clotho_captions_development.csv  
clotho_metadata_development.csv  
clotho_metadata_evaluation.csv`

and these directories (among others):

clotho_dataset_dev/
clotho_dataset_eva/  

In those directories, there are the npy files containing the log-MEL features. To generate those, you need to process the WAV files with the scripts provided in this repository:

https://github.com/audio-captioning/clotho-dataset

In main_train.py, two splits are used:

• clotho_dataset_dev for training
• clotho_dataset_eva for validation.

In main_decode.py, you can use a trained model to generate captions on:

• clotho_dataset_eva
• clotho_dataset_test, which is the official evaluation subset. No ground-truth is available for this one.