multilingual_kws

multilingual_kws copied to clipboard

TFDS api

2

mmaz

words greater than 2^3 are probably > 1s

1

mmaz

generate a sibling dataset with speech context

1

mmaz

Some percentage of wavs (~3%) are below 1s according to soxi, others can't be opened

1

mmaz

Changes: - `EDA` folder which has scripts for, well you guessed it, Exploratory Data Analysis - `extraction` has consolidated scripts for extraction. - `packaging` has scripts used for tarring, uploading,...

Sharad24

Ciroye

Sharad24

mmaz

* this verification should only apply to words that have > the minimum number of clips for defining a split * currently the tarballs for audio has the full path...

mmaz

Most of our current alignments are for Common Voice 3/4, so re-running the alignments should create a lot more data. Low priority as of now.

Sharad24