| | |
|---|
| | | return feats_pad, feats_lens, lfr_splice_frame_idxs |
|---|
| | | |
|---|
| | | def forward( |
|---|
| | | self, input: torch.Tensor, input_lengths: torch.Tensor, is_final: bool = False |
|---|
| | | self, input: torch.Tensor, input_lengths: torch.Tensor, is_final: bool = False, reset: bool = False |
|---|
| | | ) -> Tuple[torch.Tensor, torch.Tensor]: |
|---|
| | | if reset: |
|---|
| | | self.cache_reset() |
|---|
| | | batch_size = input.shape[0] |
|---|
| | | assert batch_size == 1, 'we support to extract feature online only when the batch size is equal to 1 now' |
|---|
| | | waveforms, feats, feats_lengths = self.forward_fbank(input, input_lengths) # input shape: B T D |
|---|
| | |
|---|
| | | feats_pad = pad_sequence(feats, |
|---|
| | | batch_first=True, |
|---|
| | | padding_value=0.0) |
|---|
| | | return feats_pad, feats_lens |
|---|
| | | return feats_pad, feats_lens |
|---|
