Hugging Face's logo

espnet
/
owsm_v3.1_ebf

Automatic Speech Recognition
ESPnet
multilingual
audio
speech-translation
Model card Files Community
2

Kabyle WER/CER

#2
by ayymen - opened
Discussion
ayymen

Thank you for open-sourcing this amazing project!
I just wanted to know what WER/CER you achieved on Common Voice for the Kabyle (kab) subset. I couldn't find the info in the paper.

pyf98
 ESPnet org

Hi, thanks for the question! We haven't tested on many low-resource languages, so I'm not sure about the performance of that specific data. If that test data is ready, you could probably try running inference and check the result.

ayymen

Thanks for responding! Which CV version is the model trained on? I think CV splits the data differently between versions, so I need to know the exact one.

pyf98
 ESPnet org

We used this version: https://github.com/espnet/espnet/blob/1c1a12d2c7f613ac59cc996aad8777a8456e9e95/egs2/owsm_v3/s2t1/local/prepare_commonvoice.sh#L49

ayymen

Thank you!

ayymen changed discussion status to closed

Sign up or log in to comment