Spaces:
Running on CPU Upgrade
Request to add french telephony / call-center datasets
Hello,
I've been curating two french telephony datasets.
I believe they could be a nice addition as they cover a very popular yet challenging use case, and to the best of my knowledge they are the only ones available in French.
ESLO, Accueil UBS
They come from French universities and their ground truth add some "quirks" related to anonymisation and specific tags which made them unsuitable to calculate WER.
So I processed their transcripts with Gemini Flash to clean the tags and ElevenLabs ASR to fix some rare inconsistencies with anonymisation and I manually checked the results.
What do you think ?
thanks @BrunoHays for the request! it could indeed be interest to add to the French evals. Could you open a PR to our GitHub so we can iterate on this? You can see some pointers here for adding a new language/dataset: https://github.com/huggingface/open_asr_leaderboard/issues/241