search

vocaliodmiku / wav2vec2mdd

End-to-End Mispronunciation Detection via wav2vec2.0
38 stars 6 forks source link

pre-training or finetuning? #3

Closed MengjieQian closed 2 years ago

MengjieQian commented 2 years ago

Hello,

Thanks for sharing your work with us. I have one question about your configure file, is the configure file for fine-tuning a pre-trained model? Why it is "_name: audio_pretraining" in the task session in the configure file? Also, does it work for any pre-trained Wav2Vec models? Thanks.

Best, Mengjie

vocaliodmiku commented 2 years ago

Oh, this is an old issue about fairseq 0.1.* version. the developer didn't separate finetuning from pretraining in the old version. According the up-to-date version of fairseq, you should use "task._name=audio_finetune". Good luck for u experiment.