Outliers in lingala asr lin_asr
#25
by madova - opened
Hi Waxal team,
I was reviewing the data in the Lingala subset (lin_asr) and spotted a bunch of outliers. Just to be absolutely sure, I even went back and listened to the current audios from "waxal", and they definitely still have the same anomalies.
Filtering them out is actually pretty straightforward. Just like in a previous discussion regarding other languages, the issue comes from 1-second audios that have an abnormally high words-per-second rate.
I can't attach my CSV file here, but applying this simple filter on your end should easily help you track them down.
Could you take a look when you have a moment so these can be cleaned up in a future update?
Thanks for your help and for the awesome work on this dataset!
