search

rst0070 / Rawformer-implementation-anti-spoofing

Pytorch implementation of "LEVERAGING POSITIONAL-RELATED LOCAL-GLOBAL DEPENDENCY FOR SYNTHETIC SPEECH DETECTION"
18 stars 0 forks source link

Test with variable length utterance #2

Open hungdinhxuan opened 4 months ago

hungdinhxuan commented 4 months ago

Hello @rst0070, I have re-trained the model using your code, and it worked pretty well with fixed-length utterances; however, when trying to test with variable-length utterances by adding the setting self.skip_adjustDuration = True in ASVspoof2021LA_eval:

    def __getitem__(self, index: Any) -> Any:

        utter, _, label = self.data_list[index]
        utter, _ = torchaudio.load(utter)
        If not self.skip_adjustDuration:
            utter = self.adjust duration(utter)

        Return utter, label

It raised an error like: RuntimeError: Trying to resize storage that is not resizable.

I need to learn an exact way to test with variable-length utterances. Could you help me? Thank you for your work.

rst0070 commented 4 months ago

Hello @rst0070, I have re-trained the model using your code, and it worked pretty well with fixed-length utterances; however, when trying to test with variable-length utterances by adding the setting self.skip_adjustDuration = True in ASVspoof2021LA_eval:

    def __getitem__(self, index: Any) -> Any:

        utter, _, label = self.data_list[index]
        utter, _ = torchaudio.load(utter)
        If not self.skip_adjustDuration:
            utter = self.adjust duration(utter)

        Return utter, label

It raised an error like: RuntimeError: Trying to resize storage that is not resizable.

I need to learn an exact way to test with variable-length utterances. Could you help me? Thank you for your work.

Thank you for your comment! I will try to fix the issue. However, it would take some time as I don't have access to the dataset and environment for the experiment right now.