speech emotion recognition from log-Mel spectrogram using vertically long patch

This repo is the official implementation of "Accuracy Enhancement Method for Speech Emotion Recognition from Spectrogram using Temporal Frequency Correlation and Positional Information Learning through Knowledge Transfer".

How to load model

I saved the model as python dict() format like below:

torch.save({
            'model_state_dict': model.state_dict(),
            'CE': cross_entropy_loss
            'L1': L1_loss
            ...
            }, PATH)

So, you can load the pretrained weight like below:

# model.load_state_dict(torch.load(PATH_WEIGHT_FILE)['model_state_dict'])
model.load_state_dict(torch.load('./weight/teacher_92.64_CREMA_D.ckpt')['model_state_dict'])