Hello, I made a few modifications, mainly putting stft and istft outside the forward function. The following is the modification of the "train_step" function in the "lightning_modules.py" script.
Of course, corresponding modifications have been made in "separator.py" and other scripts, but when I trained, all SDR values obtained by evaluate were 0. as follows:
I've searched for a long time but can't find the reason. Hope to get some knowledge from you, thank you.
Hello, I made a few modifications, mainly putting stft and istft outside the forward function. The following is the modification of the "train_step" function in the "lightning_modules.py" script.
` def training_step(self, batch_data: Tuple, batch_idx: int) -> torch.float: input_data, target_data = self.batch_data_preprocessor(batch_data) batch_size, channels_num, segment_samples = input_data.shape
`
Of course, corresponding modifications have been made in "separator.py" and other scripts, but when I trained, all SDR values obtained by evaluate were 0. as follows:
I've searched for a long time but can't find the reason. Hope to get some knowledge from you, thank you.