Closed leijue222 closed 2 years ago
Hi @leijue222 ,
The recipe for SUPERB evaluation can be found in s3prl repo, and the recipe for ASR fine-tuning can be found in fairseq repo. And the recipes for other downstream tasks are currently not available.
Is the recipes in s3prl?
I see unispeech_sat and wavlm in s3prl, and do not find any train/test/infer code in this repository.
WavLM SV demo in s3prl:
python3 run_downstream.py -n wavlm -m train -u fbank -d sv_voxceleb1
python3 run_downstream.py -m evaluate -e result/downstream/wavlm/wavlm_large_finetune.pth
https://github.com/microsoft/UniSpeech/tree/main/downstreams/speaker_verification#example