search

youngeun1209 / NeuroTalk

MIT License
44 stars 6 forks source link

DTW #9

Open hanbuck30 opened 2 months ago

hanbuck30 commented 2 months ago

Thank you for your code share. If you use DTW align to evaluate the output and target during evaluation, wouldn’t that mean the target is always required when assessing performance through eval.py?

youngeun1209 commented 2 months ago

I believe the code could be confusing. DTW is used only to calculate the loss, so it is necessary to evaluate the dataset. This is why DTW is included in the evaluation process. However, DTW is not required when generating the mel-spectrogram. (the optimal goal) I’ve updated the code to remove DTW when generating the mel spectrogram. Thank you for your interest!