French equivalent to MNIST Audio Dataset
Executing ./script.sh will enhance the dataset with pitch-shifted and noisy samples, also uniting resulting audio and adding the /voices audio used as an unknown class, and generate a .csv file for using in Tensorflow
You will also find utilitary scripts like slice.sh that will convert all .mp3 files in /raw to an equivalent .wav and then isolate unique digits and output each digits of the raw audio with the right naming convention. It is usefull when adding a new speaker to the dataset, after following the instructions of Guide_enregistrement.pdf.