emotion prob extract

[JingRH]

If I want to obtain the probabilities of 8 emotions:

"<|HAPPY|>",
"<|SAD|>",
"<|ANGRY|>",
"<|NEUTRAL|>",
"<|FEARFUL|>",
"<|DISGUSTED|>",
"<|SURPRISED|>",
"<|OTHER|>",
should I take the logits at the corresponding token ID positions from the second frame?

    for i in range(b):
        x = ctc_logits[i, : encoder_out_lens[i].item(), :]
        yseq = x.argmax(dim=-1)

        
        emotion_logits=x[1,:]

Can I do this?