def raw_iter()

in datasets.py [0:0]

• 21 lines of code
• 6 McCabe index (conditional complexity)

    def raw_iter(self) -> Iterator[Tuple[torch.Tensor, torch.Tensor]]:
        """
        Iterate over the dataset without extracting clips or batching.

        Yields:
          Samples from the dataset. The first element is the spectrograms and
          the second is the waveforms.

          Spectrogram shape: [num_bands, num_frames]
          Waveform shape: [num_frames * hop_samples]
        """
        # Split data among workers.
        worker_info = torch.utils.data.get_worker_info()
        if worker_info is not None:
            indices = [
                idx
                for i, idx in enumerate(self.indices)
                if i % worker_info.num_workers  # pylint: disable=no-member
                == worker_info.id  # pylint: disable=no-member
            ]
        else:
            indices = self.indices

        for key in indices:
            waveform, sr = self.dataset[key][:2]

            # Resample the waveform to a fixed sample rate
            waveform = librosa.resample(waveform[0].numpy(), sr, AUDIO_SAMPLE_RATE)
            waveform = torch.clip(torch.tensor([waveform]), -1, 1)

            # Pad to make sure waveform is a multiple of hop length.
            padding = (
                MEL_HOP_SAMPLES - waveform.numel() % MEL_HOP_SAMPLES
            ) % MEL_HOP_SAMPLES
            # pyre-fixme[6]: Expected `List[int]` for 2nd param but got `Tuple[int,
            #  typing.Any]`.
            waveform = torch.nn.functional.pad(waveform, (0, padding))

            # Compute spectrogram of waveform.
            # Length of spectrogram is exactly waveform length over hop length.
            spectrogram = self.mel(waveform)

            yield spectrogram.squeeze(0), waveform.squeeze(0)