This program does what datasets does. When you push dataset created by the audiofolder script, it creates parquet data and shard them internally.

So, you can use audios-to-dataset instead if you need faster speeds than datasets provides.

Time to podman my songs