A question

[francqz31]

Hey Author , Thanks for the opensource

I wanted to ask if emotion2vec is better than https://github.com/audeering/w2v2-how-to

Thanks in advance

[ddlBoJack]

Hi, it is worth noting that emotion2vec is a universal emotional representation model, which needs to be trained on a certain downstream task to compare the performance. For performance on a specific dataset, please refer to our paper: https://arxiv.org/abs/2312.15185
We will provide a finetuned model with massive labeled data soon, welcome to follow up and compare performance :)

[francqz31]

@ddlBoJack Oh Thanks , Noted👍. so this finetuned model with massive labeled data are you planning on training it in a way to beat "w2v2-how-to" ?
in "w2v2-how-to" They finetuned the pre-trained wav2vec2-large-robust model on MSP-Podcast (v1.7). The pre-trained model was pruned from 24 to 12 transformer layers before fine-tuning. !!

[ddlBoJack]

Our emotion2vec only has 12 layers of 768-dim transformer. As for the performance, researchers can explore further after we release the finetuned checkpoint.