MiniLibriMix dataset

MiniLibriMix is a small version of LibriMix. 

It was made for demonstration purposes. 

It contains a train set of 800 mixtures and a validation set of 200 mixtures.

In each set, you will find :

• mix_clean a folder containing clean mixtures of 2 speakers.
• mix_both a folder containing clean mixtures of 2 speakers and a noise.
• s1, s2, noise three folders containing the raw signals in the mixture.

We also provide metadata that are useful to load the data easily.

You should check out Asteroid where models, dataloaders and more resources are available.