Open Darcy0218 opened 10 months ago
Hi, since the original checkpoint of HTS-AT is trained on 32kHz data, when I evaluated the finetuning result on ESC-50, I first resampled ESC-50 data into 32kHz.
So this audio is referred to the 1-100032-A-0.wav audio file in 32kHz. You need to download the ESC-50 data and resample them to 32kHz. You can refer to torchaudio or librosa library to perform this
I think there is an error in the resampling process, I am encountering the same error
while preparing dataset for esc-50