Fileformat in ogg?

[WikiLucas00]

In lingua-libre/operations/create_datasets.sh, we can read on line 37, 43 and 52 that the fileformat asked is ogg (while the chosen format for Lingua Libre files on Commons is wav).
Is there a reason for using this format in the datasets, and is there a reason preventing us from changing it for wav format?

All the best

[Poslovitch]

As far as I know, ogg tends to be far smaller than wav (it's a format that makes use of dynamic compression, while wav is not a compressed format). Considering that the backend server has limited disk space and resources (incl. bandwidth) as most of it must be assigned to an ever-increasingly-resource-intensive BlazeGraph, we can't afford downloading and storing hundreds of GB of recordings.

IMHO, the current way datasets work has already become unsustainable. Switching from ogg to wav is not going to help 😕 .

[hugolpz]

Context

CBR stands for constant bitrate and is an encoding method that keeps the bitrate the same all along the file.
VBR, by contrast, is a variable bitrate. Bitrate varies within the file along time.

Issues

• In web pages, audio play and JS, VBR can cause miscalculation of the audio's duration, by example.
• In NLP, VBR can cause issues.