search

Jiaxin-Ye / Emo-DNA

[ACM MM 2023] Official PyTorch implementation of "Emo-DNA: Emotion Decoupling and Alignment Learning for Cross-Corpus Speech Emotion Recognition".
10 stars 1 forks source link

pseudo-label assignment #2

Open TaoRainLover opened 10 months ago

TaoRainLover commented 10 months ago

Hello author, I have a question regarding the pseudo-label assignment in implementing this paper. There are two types of labels used in the experiment (Arousal & Valence). I would like to know if the pseudo-labels assigned to the target dataset are based on the original source dataset's classification labels or on the experimental division labels (Arousal & Valence).

Looking forward to your reply!

Jiaxin-Ye commented 10 months ago

Thank you for your question. Actually, the original source datasets don't achieve dimensional labels but with categorical ones. In order to utilize all speech samples, we follow a common theoretical emotional mapping Geneva Emotion Wheel to map categorical emotion to dimensional one. Hence, we assign the dimensional pseudo-labels (i.e., arousal or valence emotion) to the unlabeled target corpus instead of categorical ones.

Wish I have solved your problem.

TaoRainLover commented 10 months ago

Thank you for your prompt and detailed reply.

So, based on the dimension labels, what I want to clarify is whether only one type of label (Arousal or Valence) is assigned to the target dataset in one experiment, or both types of labels (Arousal and Valence) are assigned simultaneously? If there is only one type, then for testing and validation purposes, should two separate experiments be conducted to determine the accuracy of the two labels assigned to the target dataset respectively?

and looking forward to your reply again!

2023年12月26日 14:37，Jiaxin Ye @.***> 写道：

Thank you for your question. Actually, the original source datasets don't achieve dimensional labels but with categorical ones. In order to utilize all speech samples, we follow a common theoretical emotional mapping Geneva Emotion Wheel to map categorical emotion to dimensional one. Hence, we assign the dimensional pseudo-labels (i.e., arousal or valence emotion) to the unlabeled target corpus instead of categorical ones.

Wish I have solved your problem.

— Reply to this email directly, view it on GitHub https://github.com/Jiaxin-Ye/Emo-DNA/issues/2#issuecomment-1869300042, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANPBK7LZFDVVNWFAWICT6MTYLJWBLAVCNFSM6AAAAABBC6JPOWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNRZGMYDAMBUGI. You are receiving this because you authored the thread.